The open alternative to Salesforce, designed for AI.
TypeScript crmcrm-systemcustomergood-first-issuegraphqlhacktoberfestjavascriptmarketingmonoreponestjsopen-sourcepostgresqlreactreactjssalestypescriptweb
feat(ai): generate every deployment catalog from the model catalog and a spec (#25831)
Félix Malfait211303ad8fea
diff --git a/.github/workflows/ci-ai-catalog-sync.yaml b/.github/workflows/ci-ai-catalog-sync.yamlindex 4b60d2bc09b..6d43d1681e7 100644--- a/.github/workflows/ci-ai-catalog-sync.yaml+++ b/.github/workflows/ci-ai-catalog-sync.yaml@@ -38,7 +38,7 @@ jobs:id: changesrun: |catalog_dir=packages/twenty-server/src/engine/metadata-modules/ai/ai-models- if git diff --quiet "$catalog_dir/ai-providers.json" "$catalog_dir/ai-model-benchmarks.json"; then+ if git diff --quiet "$catalog_dir"; thenecho "changed=false" >> "$GITHUB_OUTPUT"elseecho "changed=true" >> "$GITHUB_OUTPUT"diff --git a/packages/twenty-server/scripts/ai-catalog-sync/README.md b/packages/twenty-server/scripts/ai-catalog-sync/README.mdnew file mode 100644index 00000000000..e223da928b6--- /dev/null+++ b/packages/twenty-server/scripts/ai-catalog-sync/README.md@@ -0,0 +1,93 @@+# AI catalog++Three files in `src/engine/metadata-modules/ai/ai-models`, and two entry points+over them.++| File | What it is |+| --- | --- |+| `ai-models.json` | What every model is: identity, pricing, limits, modalities, efforts, benchmarks. No routes, no credentials. Synced daily. |+| `ai-self-host-spec.json` | What a self-hosted deployment serves: the five direct routes and their key templates. Hand-maintained. |+| `ai-providers.json` | The catalog the server bundles. Generated from the two above; a test fails if it is hand-edited. |++Cloud works the same way, from a private spec in twenty-infra, so a deployment+catalog is always a projection rather than a second copy of the truth.++## `index.ts` — sync the model catalog++Run daily by `.github/workflows/ci-ai-catalog-sync.yaml`. Reads models.dev for+model identity, pricing, context windows and availability, overlays Artificial+Analysis for intelligence, speed and cost per task, and writes `ai-models.json`+and `ai-model-benchmarks.json`, then projects the self-host spec over the first+to write `ai-providers.json`. Hand-maintained fields (`efforts`,+`dataResidency`, `zeroDataRetention`) survive the rebuild.++```bash+npx nx run twenty-server:ts-node-no-deps-transpile-only -- \+ ./scripts/ai-catalog-sync/index.ts --dry-run+```++## `project.ts` — derive a deployment catalog++A deployment serves a subset of the catalog through its own routes: Azure, a+Bedrock region, a gateway, or a first-party key. It declares only what is its+own — which routes exist, their credentials, and which catalog models each one+serves — and this projects the catalog through that spec.++```bash+npx tsx ./scripts/ai-catalog-sync/project.ts \+ --spec ./my-deployment.json \+ --out ./ai-catalog.json+```++`--catalog` points at a different `ai-models.json`; it defaults to the+committed one. The output has the same shape, so a server loads it through+`AI_CATALOG_STORAGE_PATH` with no further processing.++A spec looks like this:++```json+{+ "providers": [+ {+ "name": "azure-foundry",+ "npm": "@ai-sdk/azure",+ "label": "Azure AI Foundry",+ "apiKey": "{{AZURE_FOUNDRY_API_KEY}}",+ "baseUrl": "{{AZURE_FOUNDRY_BASE_URL}}",+ "dataResidency": "eu",+ "labelSuffix": " (Azure)",+ "models": [+ "gpt-5.6-luna",+ { "model": "gpt-5.6-sol", "cachedInputCostPerMillionTokens": 0.5 }+ ]+ },+ {+ "name": "amazon-bedrock",+ "npm": "@ai-sdk/amazon-bedrock",+ "region": "eu-central-1",+ "models": [+ { "model": "claude-opus-4-7", "as": "eu.anthropic.claude-opus-4-7" }+ ]+ }+ ]+}+```++A route that serves a whole vendor says so instead of listing its models, which+is how self-host picks up new models with the daily sync:++```json+{ "name": "openai", "npm": "@ai-sdk/openai", "models": { "vendor": "openai" } }+```++A spec can rename a model for the route that deploys it (`as`) and override the+prices it negotiated, the limits its route caps, its label and its deprecation.+It cannot restate what a model is: modalities, reasoning support, efforts and+benchmarks always come from the catalog, which is what keeps a deployment from drifting away from the+measured truth. Naming a model the catalog does not carry fails the run rather+than publishing a route to nothing.++`project.ts` resolves everything it needs by source path, so a repository+holding a private spec can run it from a sparse checkout of these files and+`packages/twenty-shared/src/ai` without installing the monorepo and without any+API key.diff --git a/packages/twenty-server/scripts/ai-catalog-sync/__tests__/carry-over-committed-fields.spec.ts b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/carry-over-committed-fields.spec.tsindex 5b4ee8466f9..88cb9ddcfbb 100644--- a/packages/twenty-server/scripts/ai-catalog-sync/__tests__/carry-over-committed-fields.spec.ts+++ b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/carry-over-committed-fields.spec.ts@@ -3,12 +3,7 @@ import { type GeneratedModel } from '../types/generated-model.type';import { carryOverCommittedFields } from '../utils/carry-over-committed-fields.util';const openai = (models: GeneratedModel[]): GeneratedCatalog => ({- openai: {- npm: '@ai-sdk/openai',- label: 'OpenAI',- apiKey: '{{OPENAI_API_KEY}}',- models,- },+ openai: { models },});describe('carryOverCommittedFields', () => {diff --git a/packages/twenty-server/scripts/ai-catalog-sync/__tests__/enrich-catalog.spec.ts b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/enrich-catalog.spec.tsindex f9c9c3eb41c..dd5447cca96 100644--- a/packages/twenty-server/scripts/ai-catalog-sync/__tests__/enrich-catalog.spec.ts+++ b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/enrich-catalog.spec.ts@@ -6,15 +6,9 @@ import { type GeneratedCatalog } from '../types/generated-catalog.type';const MEASURED_AT = '2026-09-09';-const catalogOf = (...modelNames: string[]): GeneratedCatalog =>- ({- openai: {- npm: '@ai-sdk/openai',- label: 'OpenAI',- apiKey: '',- models: modelNames.map((name) => ({ name, label: name })),- },- }) as GeneratedCatalog;+const catalogOf = (...modelNames: string[]): GeneratedCatalog => ({+ openai: { models: modelNames.map((name) => ({ name, label: name })) },+});const enrich = (catalog: GeneratedCatalog, benchmarkIndex: BenchmarkIndex) =>enrichCatalog({diff --git a/packages/twenty-server/scripts/ai-catalog-sync/__tests__/project-catalog.spec.ts b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/project-catalog.spec.tsnew file mode 100644index 00000000000..e40da6115e5--- /dev/null+++ b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/project-catalog.spec.ts@@ -0,0 +1,292 @@+import { type CatalogSpec } from '../types/catalog-spec.type';+import {+ type CanonicalCatalog,+ projectCatalog,+} from '../utils/project-catalog.util';++const canonicalCatalog: CanonicalCatalog = {+ openai: {+ npm: '@ai-sdk/openai',+ label: 'OpenAI',+ apiKey: '{{OPENAI_API_KEY}}',+ models: [+ {+ name: 'gpt-5.6-luna',+ label: 'GPT-5.6 Luna',+ inputCostPerMillionTokens: 0.2,+ outputCostPerMillionTokens: 1.2,+ cachedInputCostPerMillionTokens: 0.02,+ modalities: ['image', 'pdf'],+ efforts: ['low', 'medium', 'high'],+ benchmark: { intelligenceIndex: 37.5, measuredAt: '2026-09-11' },+ benchmarkByEffort: {+ high: { intelligenceIndex: 32.4, measuredAt: '2026-09-11' },+ },+ },+ ],+ },+ anthropic: {+ npm: '@ai-sdk/anthropic',+ label: 'Anthropic',+ apiKey: '{{ANTHROPIC_API_KEY}}',+ models: [{ name: 'claude-opus-4-7', label: 'Claude Opus 4.7' }],+ },+};++describe('projectCatalog', () => {+ it('describes a route from the catalog while keeping the route its own credentials', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ label: 'Azure AI Foundry',+ apiKey: '{{AZURE_FOUNDRY_API_KEY}}',+ baseUrl: '{{AZURE_FOUNDRY_BASE_URL}}',+ dataResidency: 'eu',+ labelSuffix: ' (Azure)',+ models: ['gpt-5.6-luna'],+ },+ ],+ };++ const projected = projectCatalog({ canonicalCatalog, spec });++ expect(Object.keys(projected)).toEqual(['azure-foundry']);+ expect(projected['azure-foundry']).toMatchObject({+ npm: '@ai-sdk/azure',+ label: 'Azure AI Foundry',+ apiKey: '{{AZURE_FOUNDRY_API_KEY}}',+ baseUrl: '{{AZURE_FOUNDRY_BASE_URL}}',+ });+ expect(projected['azure-foundry'].models?.[0]).toEqual({+ name: 'gpt-5.6-luna',+ label: 'GPT-5.6 Luna (Azure)',+ inputCostPerMillionTokens: 0.2,+ outputCostPerMillionTokens: 1.2,+ cachedInputCostPerMillionTokens: 0.02,+ modalities: ['image', 'pdf'],+ efforts: ['low', 'medium', 'high'],+ benchmark: { intelligenceIndex: 37.5, measuredAt: '2026-09-11' },+ benchmarkByEffort: {+ high: { intelligenceIndex: 32.4, measuredAt: '2026-09-11' },+ },+ dataResidency: 'eu',+ });+ });++ it('serves a model under the id the route deploys it as', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'amazon-bedrock',+ npm: '@ai-sdk/amazon-bedrock',+ region: 'eu-central-1',+ models: [+ { model: 'claude-opus-4-7', as: 'eu.anthropic.claude-opus-4-7' },+ ],+ },+ ],+ };++ const projected = projectCatalog({ canonicalCatalog, spec });++ expect(projected['amazon-bedrock'].models?.[0]).toMatchObject({+ name: 'eu.anthropic.claude-opus-4-7',+ label: 'Claude Opus 4.7',+ });+ expect(projected['amazon-bedrock'].region).toBe('eu-central-1');+ });++ it('takes a price the route negotiated and leaves the rest of the reading alone', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ models: [+ { model: 'gpt-5.6-luna', cachedInputCostPerMillionTokens: 0.5 },+ ],+ },+ ],+ };++ const projected = projectCatalog({ canonicalCatalog, spec });++ expect(projected['azure-foundry'].models?.[0]).toMatchObject({+ cachedInputCostPerMillionTokens: 0.5,+ inputCostPerMillionTokens: 0.2,+ efforts: ['low', 'medium', 'high'],+ });+ });++ it('takes a limit the route caps below the one the model allows', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'amazon-bedrock',+ npm: '@ai-sdk/amazon-bedrock',+ models: [{ model: 'gpt-5.6-luna', maxOutputTokens: 64000 }],+ },+ ],+ };++ const projected = projectCatalog({ canonicalCatalog, spec });++ expect(projected['amazon-bedrock'].models?.[0]).toMatchObject({+ maxOutputTokens: 64000,+ });+ });++ it('serves every model a vendor publishes when the spec names the vendor', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'openai',+ npm: '@ai-sdk/openai',+ apiKey: '{{OPENAI_API_KEY}}',+ models: { vendor: 'openai' },+ },+ ],+ };++ const projected = projectCatalog({ canonicalCatalog, spec });++ expect(projected['openai'].models?.map((model) => model.name)).toEqual([+ 'gpt-5.6-luna',+ ]);+ expect(projected['openai'].models?.[0]).toMatchObject({+ efforts: ['low', 'medium', 'high'],+ });+ });++ it('refuses a spec naming a vendor the catalog does not carry', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'openai',+ npm: '@ai-sdk/openai',+ models: { vendor: 'openai-imaginary' },+ },+ ],+ };++ expect(() => projectCatalog({ canonicalCatalog, spec })).toThrow(+ 'which the catalog does not carry',+ );+ });++ it('refuses a spec naming a model the catalog does not carry', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ models: ['gpt-5.6-luna', 'gpt-9-imaginary'],+ },+ ],+ };++ expect(() => projectCatalog({ canonicalCatalog, spec })).toThrow(+ 'azure-foundry/gpt-9-imaginary',+ );+ });++ it('cannot restate what a model is, only which models a route serves', () => {+ const spec = {+ providers: [+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ models: [{ model: 'gpt-5.6-luna', modalities: ['image'] }],+ },+ ],+ } as unknown as CatalogSpec;++ const projected = projectCatalog({ canonicalCatalog, spec });++ expect(projected['azure-foundry'].models?.[0]).toMatchObject({+ modalities: ['image', 'pdf'],+ });+ });++ it('refuses a repeated provider, which would drop the first one silently', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ models: ['gpt-5.6-luna'],+ },+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ models: ['claude-opus-4-7'],+ },+ ],+ };++ expect(() => projectCatalog({ canonicalCatalog, spec })).toThrow(+ 'repeated or reserved',+ );+ });++ it('refuses an SDK package or residency the server would not parse', () => {+ const withNpm = (npm: string): CatalogSpec => ({+ providers: [{ name: 'gateway', npm, models: ['gpt-5.6-luna'] }],+ });++ expect(() =>+ projectCatalog({ canonicalCatalog, spec: withNpm('@ai-sdk/imaginary') }),+ ).toThrow('unsupported SDK package');++ expect(() =>+ projectCatalog({+ canonicalCatalog,+ spec: {+ providers: [+ {+ name: 'gateway',+ npm: '@ai-sdk/openai',+ dataResidency: 'mars',+ models: ['gpt-5.6-luna'],+ },+ ],+ },+ }),+ ).toThrow('unsupported data residency');+ });++ it('refuses a negotiated price that is not a usable number', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ models: [{ model: 'gpt-5.6-luna', inputCostPerMillionTokens: -1 }],+ },+ ],+ };++ expect(() => projectCatalog({ canonicalCatalog, spec })).toThrow(+ 'which is not a usable number',+ );+ });++ it('marks the route on a label the spec overrides', () => {+ const spec: CatalogSpec = {+ providers: [+ {+ name: 'azure-foundry',+ npm: '@ai-sdk/azure',+ labelSuffix: ' (Azure)',+ models: [{ model: 'gpt-5.6-luna', label: 'Luna' }],+ },+ ],+ };++ const projected = projectCatalog({ canonicalCatalog, spec });++ expect(projected['azure-foundry'].models?.[0]?.label).toBe('Luna (Azure)');+ });+});diff --git a/packages/twenty-server/scripts/ai-catalog-sync/__tests__/self-host-projection.spec.ts b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/self-host-projection.spec.tsnew file mode 100644index 00000000000..d434686c940--- /dev/null+++ b/packages/twenty-server/scripts/ai-catalog-sync/__tests__/self-host-projection.spec.ts@@ -0,0 +1,22 @@+import canonicalCatalog from 'src/engine/metadata-modules/ai/ai-models/ai-models.json';+import shippedCatalog from 'src/engine/metadata-modules/ai/ai-models/ai-providers.json';+import selfHostSpec from 'src/engine/metadata-modules/ai/ai-models/ai-self-host-spec.json';++import { type CatalogSpec } from '../types/catalog-spec.type';+import {+ type CanonicalCatalog,+ projectCatalog,+} from '../utils/project-catalog.util';++// The shipped catalog is generated, and a hand edit to it is the drift this+// pipeline exists to remove, so it has to stay what its spec produces.+describe('the shipped catalog', () => {+ it('is what the self-host spec projects from the model catalog', () => {+ const projected = projectCatalog({+ canonicalCatalog: canonicalCatalog as CanonicalCatalog,+ spec: selfHostSpec as CatalogSpec,+ });++ expect(projected).toEqual(shippedCatalog);+ });+});diff --git a/packages/twenty-server/scripts/ai-catalog-sync/index.ts b/packages/twenty-server/scripts/ai-catalog-sync/index.tsindex c667f132e19..fd8628764c2 100644--- a/packages/twenty-server/scripts/ai-catalog-sync/index.ts+++ b/packages/twenty-server/scripts/ai-catalog-sync/index.ts@@ -12,10 +12,12 @@ import { buildCatalog } from './utils/build-catalog.util';import { carryOverCommittedFields } from './utils/carry-over-committed-fields.util';import { enrichCatalog } from './utils/enrich-catalog.util';import { fetchArtificialAnalysisBenchmarks } from './utils/fetch-artificial-analysis-benchmarks.util';+import { projectCatalog } from './utils/project-catalog.util';import { readCommittedBenchmarks } from './utils/read-committed-benchmarks.util';import { buildCoverageReport } from './utils/build-coverage-report.util';import { renderCoverageReport } from './utils/render-coverage-report.util';import { type BenchmarkIndex } from './types/benchmark-index.type';+import { type CatalogSpec } from './types/catalog-spec.type';import { type GeneratedCatalog } from './types/generated-catalog.type';const AI_MODELS_DIR = path.resolve(@@ -29,6 +31,8 @@ const AI_MODELS_DIR = path.resolve('ai-models',);+const MODELS_PATH = path.join(AI_MODELS_DIR, 'ai-models.json');+const SELF_HOST_SPEC_PATH = path.join(AI_MODELS_DIR, 'ai-self-host-spec.json');const CATALOG_PATH = path.join(AI_MODELS_DIR, 'ai-providers.json');const BENCHMARKS_PATH = path.join(AI_MODELS_DIR, 'ai-model-benchmarks.json');@@ -43,7 +47,7 @@ const readArgument = (flag: string): string | undefined => {return index === -1 ? undefined : process.argv[index + 1];};-const readCommittedCatalog = (filePath: string): GeneratedCatalog =>+const readCommittedModels = (filePath: string): GeneratedCatalog =>fs.existsSync(filePath)? (JSON.parse(fs.readFileSync(filePath, 'utf-8')) as GeneratedCatalog): {};@@ -131,7 +135,7 @@ const main = async (): Promise<void> => {carryOverCommittedFields({catalog,- committedCatalog: readCommittedCatalog(CATALOG_PATH),+ committedCatalog: readCommittedModels(MODELS_PATH),});const overlay = enrichCatalog({@@ -155,7 +159,18 @@ const main = async (): Promise<void> => {return;}- await writeJson(CATALOG_PATH, catalog);+ await writeJson(MODELS_PATH, catalog);+ // Self-host runs the same projection cloud does, from a spec that names the+ // five direct routes and lets each serve its whole vendor.+ await writeJson(+ CATALOG_PATH,+ projectCatalog({+ canonicalCatalog: catalog,+ spec: JSON.parse(+ fs.readFileSync(SELF_HOST_SPEC_PATH, 'utf-8'),+ ) as CatalogSpec,+ }),+ );await writeJson(BENCHMARKS_PATH, {source: 'artificialanalysis.ai',measuredAt,diff --git a/packages/twenty-server/scripts/ai-catalog-sync/project.ts b/packages/twenty-server/scripts/ai-catalog-sync/project.tsnew file mode 100644index 00000000000..26adbecc9b9--- /dev/null+++ b/packages/twenty-server/scripts/ai-catalog-sync/project.ts@@ -0,0 +1,134 @@+// Projects the canonical catalog through a deployment spec. See README.md.+//+// This entry point and the files it imports resolve through node builtins and+// source paths only, so a repository holding a private spec can run them+// straight from a sparse checkout without installing the monorepo.++import { mkdirSync, readFileSync, writeFileSync } from 'node:fs';+import { dirname, resolve } from 'node:path';++import { type CatalogSpec } from './types/catalog-spec.type';+import {+ type CanonicalCatalog,+ projectCatalog,+} from './utils/project-catalog.util';++const DEFAULT_CATALOG_PATH = resolve(+ __dirname,+ '..',+ '..',+ 'src',+ 'engine',+ 'metadata-modules',+ 'ai',+ 'ai-models',+ 'ai-models.json',+);++// A flag whose value is missing would otherwise swallow the next flag and write+// the catalog to a file named after it.+const readArgument = (flag: string): string | undefined => {+ const index = process.argv.indexOf(flag);++ if (index === -1) {+ return undefined;+ }++ const value = process.argv[index + 1];++ if (value === undefined || value.startsWith('--')) {+ throw new Error(`${flag} is missing its value`);+ }++ return value;+};++const readJson = <TValue>(path: string): TValue => {+ try {+ return JSON.parse(readFileSync(path, 'utf-8')) as TValue;+ } catch (error) {+ throw new Error(+ `Could not read ${path}: ${error instanceof Error ? error.message : String(error)}`,+ );+ }+};++const assertSpecIsUsable = ({+ spec,+ path,+}: {+ spec: CatalogSpec;+ path: string;+}): void => {+ if (!Array.isArray(spec.providers) || spec.providers.length === 0) {+ throw new Error(`${path} declares no providers`);+ }++ for (const provider of spec.providers) {+ if (typeof provider.name !== 'string' || provider.name.length === 0) {+ throw new Error(`${path} has a provider without a name`);+ }++ if (typeof provider.npm !== 'string' || provider.npm.length === 0) {+ throw new Error(+ `${path}: provider "${provider.name}" has no npm package`,+ );+ }++ const servesModels = Array.isArray(provider.models)+ ? provider.models.length > 0+ : typeof provider.models?.vendor === 'string' &&+ provider.models.vendor.length > 0;++ if (!servesModels) {+ throw new Error(`${path}: provider "${provider.name}" serves no models`);+ }+ }+};++const describe = (catalog: CanonicalCatalog): string => {+ const models = Object.values(catalog).flatMap(+ (provider) => provider.models ?? [],+ );+ const withEfforts = models.filter((model) => model.efforts !== undefined);+ const withBenchmark = models.filter((model) => model.benchmark !== undefined);++ return `${Object.keys(catalog).length} providers, ${models.length} models, ${withEfforts.length} with efforts, ${withBenchmark.length} with a benchmark`;+};++const main = (): void => {+ const specPath = readArgument('--spec');+ const outPath = readArgument('--out');++ if (specPath === undefined || outPath === undefined) {+ throw new Error(+ 'Usage: project.ts --spec <spec.json> --out <catalog.json> [--catalog <ai-models.json>]',+ );+ }++ const catalogPath = readArgument('--catalog') ?? DEFAULT_CATALOG_PATH;+ const spec = readJson<CatalogSpec>(specPath);++ assertSpecIsUsable({ spec, path: specPath });++ const projected = projectCatalog({+ canonicalCatalog: readJson<CanonicalCatalog>(catalogPath),+ spec,+ });++ mkdirSync(dirname(resolve(outPath)), { recursive: true });+ writeFileSync(outPath, `${JSON.stringify(projected, null, 2)}\n`, 'utf-8');++ // oxlint-disable-next-line no-console+ console.log(`Wrote ${outPath}: ${describe(projected)}`);+};++try {+ main();+} catch (error) {+ // oxlint-disable-next-line no-console+ console.error(+ `AI catalog projection failed: ${error instanceof Error ? error.message : String(error)}`,+ );+ process.exit(1);+}diff --git a/packages/twenty-server/scripts/ai-catalog-sync/types/catalog-spec.type.ts b/packages/twenty-server/scripts/ai-catalog-sync/types/catalog-spec.type.tsnew file mode 100644index 00000000000..489833d9d5f--- /dev/null+++ b/packages/twenty-server/scripts/ai-catalog-sync/types/catalog-spec.type.ts@@ -0,0 +1,46 @@+// Deliberately not extensible: a field a route cannot restate is a field it+// cannot drift on, which is the property this pipeline is for.+export type CatalogSpecModelOverrides = {+ label?: string;+ isDeprecated?: boolean;+ contextWindowTokens?: number;+ maxOutputTokens?: number;+ inputCostPerMillionTokens?: number;+ outputCostPerMillionTokens?: number;+ cachedInputCostPerMillionTokens?: number;+ cacheCreationCostPerMillionTokens?: number;+};++export type CatalogSpecModel = CatalogSpecModelOverrides & {+ model: string;+ // `eu.anthropic.claude-opus-4-7` for a Bedrock deployment of+ // `claude-opus-4-7`.+ as?: string;+};++// Every model the vendor publishes, so a route that serves a whole vendor+// picks up new models with the daily sync instead of waiting for an edit here.+export type CatalogSpecModelSelector = {+ vendor: string;+};++export type CatalogSpecProvider = {+ name: string;+ npm: string;+ label?: string;+ apiKey?: string;+ baseUrl?: string;+ region?: string;+ authType?: string;+ accessKeyId?: string;+ secretAccessKey?: string;+ sessionToken?: string;+ dataResidency?: string;+ zeroDataRetention?: boolean;+ labelSuffix?: string;+ models: (string | CatalogSpecModel)[] | CatalogSpecModelSelector;+};++export type CatalogSpec = {+ providers: CatalogSpecProvider[];+};diff --git a/packages/twenty-server/scripts/ai-catalog-sync/types/generated-provider.type.ts b/packages/twenty-server/scripts/ai-catalog-sync/types/generated-provider.type.tsindex ebc6826d505..1d8aecbb5ab 100644--- a/packages/twenty-server/scripts/ai-catalog-sync/types/generated-provider.type.ts+++ b/packages/twenty-server/scripts/ai-catalog-sync/types/generated-provider.type.ts@@ -1,10 +1,5 @@-import { type AiSdkPackage } from 'twenty-shared/ai';-import { type GeneratedModel } from './generated-model.type';export type GeneratedProvider = {- npm: AiSdkPackage;- label: string;- apiKey: string;models: GeneratedModel[];};diff --git a/packages/twenty-server/scripts/ai-catalog-sync/utils/build-catalog.util.ts b/packages/twenty-server/scripts/ai-catalog-sync/utils/build-catalog.util.tsindex b5ccdf8ac6e..a8dc74bfc46 100644--- a/packages/twenty-server/scripts/ai-catalog-sync/utils/build-catalog.util.ts+++ b/packages/twenty-server/scripts/ai-catalog-sync/utils/build-catalog.util.ts@@ -1,7 +1,4 @@-import {- type AiSdkPackage,- NATIVE_AI_SDK_PROVIDER_IDS,-} from 'twenty-shared/ai';+import { NATIVE_AI_SDK_PROVIDER_IDS } from 'twenty-shared/ai';import { isDefined } from 'twenty-shared/utils';import { type ModelsDevData } from 'src/engine/metadata-modules/ai/ai-models/types/models-dev-data.type';@@ -37,22 +34,6 @@ const EXCLUDED_MODEL_SUFFIXES = ['-audio-preview', '-realtime-preview'];const LONG_CONTEXT_THRESHOLD_TOKENS = 200000;-const PROVIDER_LABELS: Record<string, string> = {- openai: 'OpenAI',- anthropic: 'Anthropic',- google: 'Google',- mistral: 'Mistral',- xai: 'xAI',-};--const API_KEY_TEMPLATES: Record<string, string> = {- openai: '{{OPENAI_API_KEY}}',- anthropic: '{{ANTHROPIC_API_KEY}}',- google: '{{GOOGLE_API_KEY}}',- mistral: '{{MISTRAL_API_KEY}}',- xai: '{{XAI_API_KEY}}',-};-const isLanguageModel = (modelId: string): boolean => {const lowerId = modelId.toLowerCase();@@ -169,12 +150,7 @@ export const buildCatalog = (data: ModelsDevData): GeneratedCatalog => {continue;}- catalog[providerName] = {- npm: `@ai-sdk/${providerName}` as AiSdkPackage,- label: PROVIDER_LABELS[providerName] ?? providerName,- apiKey: API_KEY_TEMPLATES[providerName] ?? '',- models,- };+ catalog[providerName] = { models };}return catalog;diff --git a/packages/twenty-server/scripts/ai-catalog-sync/utils/project-catalog.util.ts b/packages/twenty-server/scripts/ai-catalog-sync/utils/project-catalog.util.tsnew file mode 100644index 00000000000..51011df4c4b--- /dev/null+++ b/packages/twenty-server/scripts/ai-catalog-sync/utils/project-catalog.util.ts@@ -0,0 +1,242 @@+// twenty-shared by source path rather than by package name: the consuming+// repository runs this from a sparse checkout with no workspace install, so+// 'twenty-shared/ai' would not resolve there.+import { isAiSdkPackage } from '../../../../twenty-shared/src/ai/utils/is-ai-sdk-package.util';+import { isDataResidency } from '../../../../twenty-shared/src/ai/utils/is-data-residency.util';++import {+ type CatalogSpec,+ type CatalogSpecModel,+ type CatalogSpecModelOverrides,+ type CatalogSpecProvider,+} from '../types/catalog-spec.type';++// Structural, so the catalog needs no schema here and this file can run from a+// checkout that has not installed the monorepo.+type CanonicalModel = { name: string; label?: string } & Record<+ string,+ unknown+>;+type CanonicalProvider = { models?: CanonicalModel[] } & Record<+ string,+ unknown+>;+export type CanonicalCatalog = Record<string, CanonicalProvider>;++const OVERRIDABLE_FIELDS = [+ 'label',+ 'isDeprecated',+ 'contextWindowTokens',+ 'maxOutputTokens',+ 'inputCostPerMillionTokens',+ 'outputCostPerMillionTokens',+ 'cachedInputCostPerMillionTokens',+ 'cacheCreationCostPerMillionTokens',+] as const satisfies readonly (keyof CatalogSpecModelOverrides)[];++const NUMERIC_FIELDS = [+ 'contextWindowTokens',+ 'maxOutputTokens',+ 'inputCostPerMillionTokens',+ 'outputCostPerMillionTokens',+ 'cachedInputCostPerMillionTokens',+ 'cacheCreationCostPerMillionTokens',+] as const satisfies readonly (keyof CatalogSpecModelOverrides)[];++const PROVIDER_CREDENTIAL_FIELDS = [+ 'label',+ 'apiKey',+ 'baseUrl',+ 'region',+ 'authType',+ 'accessKeyId',+ 'secretAccessKey',+ 'sessionToken',+] as const;++const asSpecModel = (entry: string | CatalogSpecModel): CatalogSpecModel =>+ typeof entry === 'string' ? { model: entry } : entry;++const indexCanonicalModels = (+ catalog: CanonicalCatalog,+): Map<string, CanonicalModel> => {+ const models = new Map<string, CanonicalModel>();++ for (const provider of Object.values(catalog)) {+ for (const model of provider.models ?? []) {+ models.set(model.name, model);+ }+ }++ return models;+};++const projectModel = ({+ canonicalModel,+ specModel,+ provider,+}: {+ canonicalModel: CanonicalModel;+ specModel: CatalogSpecModel;+ provider: CatalogSpecProvider;+}): CanonicalModel => {+ const overrides = Object.fromEntries(+ OVERRIDABLE_FIELDS.filter((field) => specModel[field] !== undefined).map(+ (field) => [field, specModel[field]],+ ),+ );++ const routeFields = {+ ...(provider.dataResidency === undefined+ ? {}+ : { dataResidency: provider.dataResidency }),+ ...(provider.zeroDataRetention === undefined+ ? {}+ : { zeroDataRetention: provider.zeroDataRetention }),+ };++ return {+ ...canonicalModel,+ ...routeFields,+ ...overrides,+ name: specModel.as ?? canonicalModel.name,+ // The suffix marks the route, so it is appended to whichever name the model+ // ends up with rather than to the catalog's.+ label: `${specModel.label ?? canonicalModel.label ?? canonicalModel.name}${provider.labelSuffix ?? ''}`,+ };+};++const assertProviderIsUsable = ({+ provider,+ seenNames,+}: {+ provider: CatalogSpecProvider;+ seenNames: Set<string>;+}): void => {+ // Providers key an object, so a repeat would drop the first one's credentials+ // and models on the floor, and `__proto__` would drop its own.+ if (seenNames.has(provider.name) || provider.name === '__proto__') {+ throw new Error(`Provider "${provider.name}" is repeated or reserved`);+ }++ if (!isAiSdkPackage(provider.npm)) {+ throw new Error(+ `Provider "${provider.name}" names an unsupported SDK package: ${provider.npm}`,+ );+ }++ if (+ provider.dataResidency !== undefined &&+ !isDataResidency(provider.dataResidency)+ ) {+ throw new Error(+ `Provider "${provider.name}" names an unsupported data residency: ${provider.dataResidency}`,+ );+ }+};++const assertNumbersAreUsable = ({+ provider,+ specModel,+}: {+ provider: CatalogSpecProvider;+ specModel: CatalogSpecModel;+}): void => {+ for (const field of NUMERIC_FIELDS) {+ const value = specModel[field];++ if (value !== undefined && (!Number.isFinite(value) || value < 0)) {+ throw new Error(+ `${provider.name}/${specModel.model} sets ${field} to ${value}, which is not a usable number`,+ );+ }+ }+};++// A spec naming a model the catalog does not carry is the failure this whole+// pipeline exists to prevent: it publishes a route to nothing, and the tier+// chains then fall through to a neighbouring rung in silence.+const resolveModels = ({+ provider,+ canonicalCatalog,+ canonicalModels,+}: {+ provider: CatalogSpecProvider;+ canonicalCatalog: CanonicalCatalog;+ canonicalModels: Map<string, CanonicalModel>;+}): { canonicalModel: CanonicalModel; specModel: CatalogSpecModel }[] => {+ if (!Array.isArray(provider.models)) {+ const { vendor } = provider.models;+ const vendorModels = canonicalCatalog[vendor]?.models ?? [];++ if (vendorModels.length === 0) {+ throw new Error(+ `Provider "${provider.name}" serves every model from "${vendor}", which the catalog does not carry`,+ );+ }++ return vendorModels.map((canonicalModel) => ({+ canonicalModel,+ specModel: { model: canonicalModel.name },+ }));+ }++ const specModels = provider.models.map(asSpecModel);+ const unknown = specModels+ .filter(({ model }) => !canonicalModels.has(model))+ .map(({ model }) => `${provider.name}/${model}`);++ if (unknown.length > 0) {+ throw new Error(+ `The catalog does not carry: ${unknown.join(', ')}. Check the model names, or wait for the catalog sync to pick them up.`,+ );+ }++ return specModels.map((specModel) => {+ assertNumbersAreUsable({ provider, specModel });++ const canonicalModel = canonicalModels.get(specModel.model);++ if (canonicalModel === undefined) {+ throw new Error(`The catalog does not carry ${specModel.model}`);+ }++ return { canonicalModel, specModel };+ });+};++export const projectCatalog = ({+ canonicalCatalog,+ spec,+}: {+ canonicalCatalog: CanonicalCatalog;+ spec: CatalogSpec;+}): CanonicalCatalog => {+ const canonicalModels = indexCanonicalModels(canonicalCatalog);+ const projected: CanonicalCatalog = {};+ const seenNames = new Set<string>();++ for (const provider of spec.providers) {+ assertProviderIsUsable({ provider, seenNames });+ seenNames.add(provider.name);++ const credentials = Object.fromEntries(+ PROVIDER_CREDENTIAL_FIELDS.filter(+ (field) => provider[field] !== undefined,+ ).map((field) => [field, provider[field]]),+ );++ projected[provider.name] = {+ npm: provider.npm,+ label: provider.label ?? provider.name,+ ...credentials,+ models: resolveModels({+ provider,+ canonicalCatalog,+ canonicalModels,+ }).map((resolved) => projectModel({ ...resolved, provider })),+ };+ }++ return projected;+};diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-models/ai-models.json b/packages/twenty-server/src/engine/metadata-modules/ai/ai-models/ai-models.jsonnew file mode 100644index 00000000000..1f044281fd5--- /dev/null+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-models/ai-models.json@@ -0,0 +1,2274 @@+{+ "openai": {+ "models": [+ {+ "name": "gpt-5-nano",+ "label": "GPT-5 Nano",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.05,+ "outputCostPerMillionTokens": 0.4,+ "cachedInputCostPerMillionTokens": 0.005,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 13,+ "outputTokensPerSecond": 150.72,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-4.1-nano",+ "label": "GPT-4.1 nano",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.1,+ "outputCostPerMillionTokens": 0.4,+ "cachedInputCostPerMillionTokens": 0.025,+ "contextWindowTokens": 1047576,+ "maxOutputTokens": 32768,+ "modalities": ["image"],+ "benchmark": {+ "intelligenceIndex": 7.8,+ "outputTokensPerSecond": 116.69,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gpt-4o-2024-05-13",+ "label": "GPT-4o (2024-05-13)",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 15,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 4096,+ "modalities": ["image"],+ "benchmark": {+ "intelligenceIndex": 7.3,+ "outputTokensPerSecond": 100.41,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gpt-5-pro",+ "label": "GPT-5 Pro",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 15,+ "outputCostPerMillionTokens": 120,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 272000,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "gpt-5.6-sol",+ "label": "GPT-5.6 Sol",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 20,+ "cachedInputCostPerMillionTokens": 0.4,+ "cacheCreationCostPerMillionTokens": 5,+ "longContextCost": {+ "inputCostPerMillionTokens": 8,+ "outputCostPerMillionTokens": 30,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.8,+ "cacheCreationCostPerMillionTokens": 10+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["none", "low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 47.1,+ "outputTokensPerSecond": 57.38,+ "costPerTask": 1.9885,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "none": {+ "intelligenceIndex": 28.3,+ "outputTokensPerSecond": 61.66,+ "effort": "none",+ "measuredAt": "2026-09-13"+ },+ "low": {+ "intelligenceIndex": 33.8,+ "outputTokensPerSecond": 56.34,+ "costPerTask": 0.2606,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 39.5,+ "outputTokensPerSecond": 56.62,+ "costPerTask": 0.505,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 42.5,+ "outputTokensPerSecond": 60.67,+ "costPerTask": 0.808,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 44.1,+ "outputTokensPerSecond": 61.12,+ "costPerTask": 1.1844,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 47.1,+ "outputTokensPerSecond": 57.38,+ "costPerTask": 1.9885,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gpt-4o-2024-08-06",+ "label": "GPT-4o (2024-08-06)",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 10,+ "cachedInputCostPerMillionTokens": 1.25,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "modalities": ["image"],+ "benchmark": {+ "intelligenceIndex": 7.7,+ "outputTokensPerSecond": 88.48,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-6-astra",+ "label": "GPT-6 Astra",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 10,+ "outputCostPerMillionTokens": 50,+ "cachedInputCostPerMillionTokens": 1,+ "cacheCreationCostPerMillionTokens": 12.5,+ "longContextCost": {+ "inputCostPerMillionTokens": 20,+ "outputCostPerMillionTokens": 75,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 2,+ "cacheCreationCostPerMillionTokens": 25+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["minimal", "low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 52.8,+ "outputTokensPerSecond": 56,+ "costPerTask": 3.2575,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 46,+ "outputTokensPerSecond": 52.11,+ "costPerTask": 0.8175,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 49.7,+ "outputTokensPerSecond": 49.98,+ "costPerTask": 1.5406,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 51,+ "outputTokensPerSecond": 49.81,+ "costPerTask": 1.7214,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 52.5,+ "outputTokensPerSecond": 51.14,+ "costPerTask": 2.3088,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 52.8,+ "outputTokensPerSecond": 56,+ "costPerTask": 3.2575,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gpt-5.2-pro",+ "label": "GPT-5.2 Pro",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 21,+ "outputCostPerMillionTokens": 168,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "gpt-5.3-codex-spark",+ "label": "GPT-5.3 Codex Spark",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.75,+ "outputCostPerMillionTokens": 14,+ "cachedInputCostPerMillionTokens": 0.175,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 32000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "gpt-4.1-mini",+ "label": "GPT-4.1 mini",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 1.6,+ "cachedInputCostPerMillionTokens": 0.1,+ "contextWindowTokens": 1047576,+ "maxOutputTokens": 32768,+ "modalities": ["image", "pdf"],+ "benchmark": {+ "intelligenceIndex": 10.2,+ "outputTokensPerSecond": 106.21,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.4",+ "label": "GPT-5.4",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 15,+ "cachedInputCostPerMillionTokens": 0.25,+ "longContextCost": {+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 22.5,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.5+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["none", "low", "medium", "high", "xhigh"],+ "benchmark": {+ "intelligenceIndex": 39,+ "outputTokensPerSecond": 131.5,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "none": {+ "intelligenceIndex": 18.2,+ "outputTokensPerSecond": 92.12,+ "effort": "none",+ "measuredAt": "2026-09-13"+ },+ "low": {+ "intelligenceIndex": 27.6,+ "outputTokensPerSecond": 90.64,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 39,+ "outputTokensPerSecond": 131.5,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gpt-4-turbo",+ "label": "GPT-4 Turbo",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 10,+ "outputCostPerMillionTokens": 30,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 4096,+ "modalities": ["image"],+ "benchmark": {+ "intelligenceIndex": 7,+ "outputTokensPerSecond": 31.1,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gpt-5.1",+ "label": "GPT-5.1",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.25,+ "outputCostPerMillionTokens": 10,+ "cachedInputCostPerMillionTokens": 0.125,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 24.7,+ "outputTokensPerSecond": 99.46,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "o1",+ "label": "o1",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 15,+ "outputCostPerMillionTokens": 60,+ "cachedInputCostPerMillionTokens": 7.5,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 100000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 15.2,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gpt-4o",+ "label": "GPT-4o",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 10,+ "cachedInputCostPerMillionTokens": 1.25,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "modalities": ["image", "pdf"],+ "benchmark": {+ "intelligenceIndex": 9,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.6-luna",+ "label": "GPT-5.6 Luna",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.2,+ "outputCostPerMillionTokens": 1.2,+ "cachedInputCostPerMillionTokens": 0.02,+ "cacheCreationCostPerMillionTokens": 0.25,+ "longContextCost": {+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 1.8,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.04,+ "cacheCreationCostPerMillionTokens": 0.5+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["none", "low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 37.5,+ "outputTokensPerSecond": 119.03,+ "costPerTask": 0.1783,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "none": {+ "intelligenceIndex": 16.8,+ "outputTokensPerSecond": 104.26,+ "effort": "none",+ "measuredAt": "2026-09-13"+ },+ "low": {+ "intelligenceIndex": 21.5,+ "outputTokensPerSecond": 105.4,+ "costPerTask": 0.0098,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 25.5,+ "outputTokensPerSecond": 104.37,+ "costPerTask": 0.0156,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 32.4,+ "outputTokensPerSecond": 101.18,+ "costPerTask": 0.044,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 34.8,+ "outputTokensPerSecond": 108.33,+ "costPerTask": 0.0853,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 37.5,+ "outputTokensPerSecond": 119.03,+ "costPerTask": 0.1783,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gpt-5.3-codex",+ "label": "GPT-5.3 Codex",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.75,+ "outputCostPerMillionTokens": 14,+ "cachedInputCostPerMillionTokens": 0.175,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 32.5,+ "outputTokensPerSecond": 127.21,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-4o-mini",+ "label": "GPT-4o mini",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.15,+ "outputCostPerMillionTokens": 0.6,+ "cachedInputCostPerMillionTokens": 0.075,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "modalities": ["image", "pdf"],+ "benchmark": {+ "intelligenceIndex": 6.7,+ "outputTokensPerSecond": 150.95,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "o1-pro",+ "label": "o1-pro",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 150,+ "outputCostPerMillionTokens": 600,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 100000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 12.4,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gpt-4.1",+ "label": "GPT-4.1",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 8,+ "cachedInputCostPerMillionTokens": 0.5,+ "contextWindowTokens": 1047576,+ "maxOutputTokens": 32768,+ "modalities": ["image", "pdf"],+ "benchmark": {+ "intelligenceIndex": 12.7,+ "outputTokensPerSecond": 151.06,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.4-nano",+ "label": "GPT-5.4 nano",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.2,+ "outputCostPerMillionTokens": 1.25,+ "cachedInputCostPerMillionTokens": 0.02,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 21.2,+ "outputTokensPerSecond": 154.16,+ "costPerTask": 0.1831,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.5-pro",+ "label": "GPT-5.5 Pro",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 30,+ "outputCostPerMillionTokens": 180,+ "longContextCost": {+ "inputCostPerMillionTokens": 60,+ "outputCostPerMillionTokens": 270,+ "thresholdTokens": 200000+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "gpt-5.4-mini",+ "label": "GPT-5.4 mini",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.75,+ "outputCostPerMillionTokens": 4.5,+ "cachedInputCostPerMillionTokens": 0.075,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 24.6,+ "outputTokensPerSecond": 186.25,+ "costPerTask": 0.4097,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.6",+ "label": "GPT-5.6",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 20,+ "cachedInputCostPerMillionTokens": 0.4,+ "cacheCreationCostPerMillionTokens": 5,+ "longContextCost": {+ "inputCostPerMillionTokens": 8,+ "outputCostPerMillionTokens": 30,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.8,+ "cacheCreationCostPerMillionTokens": 10+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "gpt-5-mini",+ "label": "GPT-5 Mini",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 0.25,+ "outputCostPerMillionTokens": 2,+ "cachedInputCostPerMillionTokens": 0.025,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 20.6,+ "outputTokensPerSecond": 86.09,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.4-pro",+ "label": "GPT-5.4 Pro",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 30,+ "outputCostPerMillionTokens": 180,+ "longContextCost": {+ "inputCostPerMillionTokens": 60,+ "outputCostPerMillionTokens": 270,+ "thresholdTokens": 200000+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "gpt-5.6-terra",+ "label": "GPT-5.6 Terra",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 12,+ "cachedInputCostPerMillionTokens": 0.2,+ "cacheCreationCostPerMillionTokens": 2.5,+ "longContextCost": {+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 18,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4,+ "cacheCreationCostPerMillionTokens": 5+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["none", "low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 42.3,+ "outputTokensPerSecond": 97.3,+ "costPerTask": 1.3987,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "none": {+ "intelligenceIndex": 22.3,+ "outputTokensPerSecond": 85.24,+ "effort": "none",+ "measuredAt": "2026-09-13"+ },+ "low": {+ "intelligenceIndex": 27.9,+ "outputTokensPerSecond": 81.07,+ "costPerTask": 0.1445,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 30.4,+ "outputTokensPerSecond": 85.38,+ "costPerTask": 0.1834,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 34.5,+ "outputTokensPerSecond": 79.06,+ "costPerTask": 0.3379,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 38.2,+ "outputTokensPerSecond": 82.58,+ "costPerTask": 0.6318,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 42.3,+ "outputTokensPerSecond": 97.3,+ "costPerTask": 1.3987,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gpt-4",+ "label": "GPT-4",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 30,+ "outputCostPerMillionTokens": 60,+ "contextWindowTokens": 8192,+ "maxOutputTokens": 8192,+ "benchmark": {+ "intelligenceIndex": 6.7,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gpt-5.2",+ "label": "GPT-5.2",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.75,+ "outputCostPerMillionTokens": 14,+ "cachedInputCostPerMillionTokens": 0.175,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 30.4,+ "outputTokensPerSecond": 64.67,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5",+ "label": "GPT-5",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.25,+ "outputCostPerMillionTokens": 10,+ "cachedInputCostPerMillionTokens": 0.125,+ "contextWindowTokens": 400000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 23,+ "outputTokensPerSecond": 90.91,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.2-chat-latest",+ "label": "GPT-5.2 Chat",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.75,+ "outputCostPerMillionTokens": 14,+ "cachedInputCostPerMillionTokens": 0.175,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "modalities": ["image"],+ "supportsReasoning": true,+ "isDeprecated": true+ },+ {+ "name": "o4-mini",+ "label": "o4-mini",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.1,+ "outputCostPerMillionTokens": 4.4,+ "cachedInputCostPerMillionTokens": 0.275,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 100000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 16.7,+ "outputTokensPerSecond": 135.62,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gpt-realtime-2.1",+ "label": "GPT-Realtime-2.1",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 24,+ "cachedInputCostPerMillionTokens": 0.4,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 32000,+ "modalities": ["audio", "image"],+ "supportsReasoning": true+ },+ {+ "name": "o3-mini",+ "label": "o3-mini",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.1,+ "outputCostPerMillionTokens": 4.4,+ "cachedInputCostPerMillionTokens": 0.55,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 100000,+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 12.5,+ "outputTokensPerSecond": 219.12,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "o3",+ "label": "o3",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 8,+ "cachedInputCostPerMillionTokens": 0.5,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 100000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 20.2,+ "outputTokensPerSecond": 107.03,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "o3-pro",+ "label": "o3-pro",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 20,+ "outputCostPerMillionTokens": 80,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 100000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 21.9,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gpt-5.3-chat-latest",+ "label": "GPT-5.3 Chat (latest)",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 1.75,+ "outputCostPerMillionTokens": 14,+ "cachedInputCostPerMillionTokens": 0.175,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "modalities": ["image"],+ "isDeprecated": true+ },+ {+ "name": "gpt-5.5",+ "label": "GPT-5.5",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 30,+ "cachedInputCostPerMillionTokens": 0.5,+ "longContextCost": {+ "inputCostPerMillionTokens": 10,+ "outputCostPerMillionTokens": 45,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 1+ },+ "contextWindowTokens": 1050000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["none", "low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 38.6,+ "outputTokensPerSecond": 79.68,+ "costPerTask": 2.634,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "none": {+ "intelligenceIndex": 23.2,+ "outputTokensPerSecond": 77.88,+ "effort": "none",+ "measuredAt": "2026-09-13"+ },+ "low": {+ "intelligenceIndex": 30.7,+ "outputTokensPerSecond": 82.33,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 34.2,+ "outputTokensPerSecond": 73.12,+ "costPerTask": 0.9017,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 37.3,+ "outputTokensPerSecond": 80.46,+ "costPerTask": 1.5413,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 38.6,+ "outputTokensPerSecond": 79.68,+ "costPerTask": 2.634,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gpt-4o-2024-11-20",+ "label": "GPT-4o (2024-11-20)",+ "modelFamily": "GPT",+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 10,+ "cachedInputCostPerMillionTokens": 1.25,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "modalities": ["image"],+ "benchmark": {+ "intelligenceIndex": 9,+ "measuredAt": "2026-09-13"+ }+ }+ ]+ },+ "anthropic": {+ "models": [+ {+ "name": "claude-sonnet-4-6",+ "label": "Claude Sonnet 4.6",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 3,+ "outputCostPerMillionTokens": 15,+ "cachedInputCostPerMillionTokens": 0.3,+ "cacheCreationCostPerMillionTokens": 3.75,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "max"],+ "benchmark": {+ "intelligenceIndex": 30.5,+ "outputTokensPerSecond": 53.65,+ "costPerTask": 2.4859,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 23.3,+ "outputTokensPerSecond": 42.24,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 24.7,+ "outputTokensPerSecond": 42,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 30.5,+ "outputTokensPerSecond": 53.65,+ "costPerTask": 2.4859,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "claude-opus-5",+ "label": "Claude Opus 5",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 25,+ "cachedInputCostPerMillionTokens": 0.5,+ "cacheCreationCostPerMillionTokens": 6.25,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 50.7,+ "outputTokensPerSecond": 50.41,+ "costPerTask": 5.8584,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 39.8,+ "outputTokensPerSecond": 47.63,+ "costPerTask": 1.0983,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 45.1,+ "outputTokensPerSecond": 48.27,+ "costPerTask": 2.1895,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 48.2,+ "outputTokensPerSecond": 50.34,+ "costPerTask": 3.6133,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 49.7,+ "outputTokensPerSecond": 49.3,+ "costPerTask": 4.8778,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 50.7,+ "outputTokensPerSecond": 50.41,+ "costPerTask": 5.8584,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "claude-opus-4-5",+ "label": "Claude Opus 4.5 (latest)",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 25,+ "cachedInputCostPerMillionTokens": 0.5,+ "cacheCreationCostPerMillionTokens": 6.25,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 64000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 29.1,+ "outputTokensPerSecond": 44.74,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "claude-fable-5-1",+ "label": "Claude Fable 5.1",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 10,+ "outputCostPerMillionTokens": 50,+ "cachedInputCostPerMillionTokens": 0.25,+ "cacheCreationCostPerMillionTokens": 12.5,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 53.4,+ "outputTokensPerSecond": 65.8,+ "costPerTask": 7.6297,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 47,+ "outputTokensPerSecond": 47.44,+ "costPerTask": 2.371,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 49.1,+ "outputTokensPerSecond": 49.63,+ "costPerTask": 2.9826,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 51.2,+ "outputTokensPerSecond": 49.2,+ "costPerTask": 3.9125,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 53.2,+ "outputTokensPerSecond": 57.28,+ "costPerTask": 5.9783,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 53.4,+ "outputTokensPerSecond": 65.8,+ "costPerTask": 7.6297,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "claude-opus-4-6",+ "label": "Claude Opus 4.6",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 25,+ "cachedInputCostPerMillionTokens": 0.5,+ "cacheCreationCostPerMillionTokens": 6.25,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "max"],+ "benchmark": {+ "intelligenceIndex": 31.9,+ "outputTokensPerSecond": 39.11,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "high": {+ "intelligenceIndex": 26.4,+ "outputTokensPerSecond": 36.56,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 31.9,+ "outputTokensPerSecond": 39.11,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "claude-sonnet-4-5-20250929",+ "label": "Claude Sonnet 4.5",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 3,+ "outputCostPerMillionTokens": 15,+ "cachedInputCostPerMillionTokens": 0.3,+ "cacheCreationCostPerMillionTokens": 3.75,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 64000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "claude-opus-4-7",+ "label": "Claude Opus 4.7",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 25,+ "cachedInputCostPerMillionTokens": 0.5,+ "cacheCreationCostPerMillionTokens": 6.25,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 40.7,+ "outputTokensPerSecond": 44.18,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "high": {+ "intelligenceIndex": 30.9,+ "outputTokensPerSecond": 43.39,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 40.7,+ "outputTokensPerSecond": 44.18,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "claude-haiku-4-5-20251001",+ "label": "Claude Haiku 4.5",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 1,+ "outputCostPerMillionTokens": 5,+ "cachedInputCostPerMillionTokens": 0.1,+ "cacheCreationCostPerMillionTokens": 1.25,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 64000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "claude-fable-5",+ "label": "Claude Fable 5",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 10,+ "outputCostPerMillionTokens": 50,+ "cachedInputCostPerMillionTokens": 1,+ "cacheCreationCostPerMillionTokens": 12.5,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 49.7,+ "outputTokensPerSecond": 66.79,+ "costPerTask": 8.746,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "max": {+ "intelligenceIndex": 49.7,+ "outputTokensPerSecond": 66.79,+ "costPerTask": 8.746,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "claude-haiku-4-5",+ "label": "Claude Haiku 4.5 (latest)",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 1,+ "outputCostPerMillionTokens": 5,+ "cachedInputCostPerMillionTokens": 0.1,+ "cacheCreationCostPerMillionTokens": 1.25,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 64000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "claude-sonnet-4-5",+ "label": "Claude Sonnet 4.5 (latest)",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 3,+ "outputCostPerMillionTokens": 15,+ "cachedInputCostPerMillionTokens": 0.3,+ "cacheCreationCostPerMillionTokens": 3.75,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 64000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "claude-opus-4-8",+ "label": "Claude Opus 4.8",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 25,+ "cachedInputCostPerMillionTokens": 0.5,+ "cacheCreationCostPerMillionTokens": 6.25,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "xhigh", "max"]+ },+ {+ "name": "claude-sonnet-5",+ "label": "Claude Sonnet 5",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 10,+ "cachedInputCostPerMillionTokens": 0.2,+ "cacheCreationCostPerMillionTokens": 2.5,+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 128000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "xhigh", "max"],+ "benchmark": {+ "intelligenceIndex": 38.4,+ "outputTokensPerSecond": 72.97,+ "costPerTask": 5.0912,+ "effort": "max",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 24.7,+ "outputTokensPerSecond": 55.84,+ "costPerTask": 0.5088,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 28.4,+ "outputTokensPerSecond": 57.81,+ "costPerTask": 0.9991,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 32,+ "outputTokensPerSecond": 58.18,+ "costPerTask": 1.7925,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 34.7,+ "outputTokensPerSecond": 62.4,+ "costPerTask": 2.872,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ },+ "max": {+ "intelligenceIndex": 38.4,+ "outputTokensPerSecond": 72.97,+ "costPerTask": 5.0912,+ "effort": "max",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "claude-opus-4-5-20251101",+ "label": "Claude Opus 4.5",+ "modelFamily": "CLAUDE",+ "inputCostPerMillionTokens": 5,+ "outputCostPerMillionTokens": 25,+ "cachedInputCostPerMillionTokens": 0.5,+ "cacheCreationCostPerMillionTokens": 6.25,+ "contextWindowTokens": 200000,+ "maxOutputTokens": 64000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 29.1,+ "outputTokensPerSecond": 44.74,+ "measuredAt": "2026-09-13"+ }+ }+ ]+ },+ "google": {+ "models": [+ {+ "name": "gemini-3.1-pro-preview-customtools",+ "label": "Gemini 3.1 Pro Preview Custom Tools",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 12,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 18,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "gemini-3.1-flash-lite-image",+ "label": "Nano Banana 2 Lite",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.25,+ "outputCostPerMillionTokens": 30,+ "contextWindowTokens": 65536,+ "maxOutputTokens": 65536,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "deep-research-max-preview-04-2026",+ "label": "Deep Research Max Preview (Apr-21-2026)",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 12,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 18,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 131072,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "gemini-3.1-pro-preview",+ "label": "Gemini 3.1 Pro Preview",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 12,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 18,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 30.4,+ "outputTokensPerSecond": 109.62,+ "costPerTask": 0.6747,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "deep-research-preview-04-2026",+ "label": "Deep Research Preview (Apr-21-2026)",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 12,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 18,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 131072,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "gemini-2.5-flash-lite",+ "label": "Gemini 2.5 Flash-Lite",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.1,+ "outputCostPerMillionTokens": 0.4,+ "cachedInputCostPerMillionTokens": 0.01,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "audio", "video", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 8.5,+ "outputTokensPerSecond": 334,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gemini-2.5-computer-use-preview-10-2025",+ "label": "Gemini 2.5 Computer Use Preview 10-2025",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 1.25,+ "outputCostPerMillionTokens": 10,+ "longContextCost": {+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 15,+ "thresholdTokens": 200000+ },+ "contextWindowTokens": 131072,+ "maxOutputTokens": 65536,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "gemini-3.6-flash",+ "label": "Gemini 3.6 Flash",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.75,+ "outputCostPerMillionTokens": 3.75,+ "cachedInputCostPerMillionTokens": 0.075,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "efforts": ["minimal", "low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 34.3,+ "outputTokensPerSecond": 195.28,+ "costPerTask": 0.9288,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "high": {+ "intelligenceIndex": 34.3,+ "outputTokensPerSecond": 195.28,+ "costPerTask": 0.9288,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gemini-3.1-flash-lite",+ "label": "Gemini 3.1 Flash Lite",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.25,+ "outputCostPerMillionTokens": 1.5,+ "cachedInputCostPerMillionTokens": 0.025,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 16,+ "outputTokensPerSecond": 287.6,+ "costPerTask": 0.0394,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gemini-3.1-flash-live-preview",+ "label": "Gemini 3.1 Flash Live Preview",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.75,+ "outputCostPerMillionTokens": 4.5,+ "contextWindowTokens": 131072,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio"],+ "supportsReasoning": true+ },+ {+ "name": "gemini-3.5-flash",+ "label": "Gemini 3.5 Flash",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 1.5,+ "outputCostPerMillionTokens": 9,+ "cachedInputCostPerMillionTokens": 0.15,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "efforts": ["minimal", "low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 33.6,+ "outputTokensPerSecond": 223.96,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "minimal": {+ "intelligenceIndex": 23.8,+ "outputTokensPerSecond": 205.07,+ "effort": "minimal",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 33.6,+ "outputTokensPerSecond": 223.96,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 33,+ "outputTokensPerSecond": 211.91,+ "costPerTask": 1.5625,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gemini-3.1-flash-lite-preview",+ "label": "Gemini 3.1 Flash Lite Preview",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.25,+ "outputCostPerMillionTokens": 1.5,+ "cachedInputCostPerMillionTokens": 0.025,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 16,+ "outputTokensPerSecond": 287.6,+ "costPerTask": 0.0394,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "gemini-3.5-flash-lite",+ "label": "Gemini 3.5 Flash Lite",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.3,+ "outputCostPerMillionTokens": 2.5,+ "cachedInputCostPerMillionTokens": 0.03,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "efforts": ["minimal", "low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 22.7,+ "outputTokensPerSecond": 320.2,+ "costPerTask": 0.1235,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gemini-flash-lite-latest",+ "label": "Gemini Flash-Lite Latest",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.3,+ "outputCostPerMillionTokens": 2.5,+ "cachedInputCostPerMillionTokens": 0.03,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 22.7,+ "outputTokensPerSecond": 320.2,+ "costPerTask": 0.1235,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gemini-3-flash-preview",+ "label": "Gemini 3 Flash Preview",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.5,+ "outputCostPerMillionTokens": 3,+ "cachedInputCostPerMillionTokens": 0.05,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "efforts": ["minimal", "low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 26.3,+ "outputTokensPerSecond": 184.98,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gemini-3.8-flash",+ "label": "Gemini 3.8 Flash",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.75,+ "outputCostPerMillionTokens": 3.75,+ "cachedInputCostPerMillionTokens": 0.075,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 41.2,+ "outputTokensPerSecond": 261.3,+ "costPerTask": 1.2428,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 33.8,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 40,+ "costPerTask": 0.931,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 41.2,+ "outputTokensPerSecond": 261.3,+ "costPerTask": 1.2428,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gemini-3.7-flash",+ "label": "Gemini 3.7 Flash",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.75,+ "outputCostPerMillionTokens": 3.75,+ "cachedInputCostPerMillionTokens": 0.075,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 39.6,+ "outputTokensPerSecond": 272.47,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 36.9,+ "outputTokensPerSecond": 273.31,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 39.6,+ "outputTokensPerSecond": 272.47,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 39.4,+ "outputTokensPerSecond": 287.8,+ "costPerTask": 0.9253,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "gemini-2.5-pro",+ "label": "Gemini 2.5 Pro",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 1.25,+ "outputCostPerMillionTokens": 10,+ "cachedInputCostPerMillionTokens": 0.125,+ "longContextCost": {+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 15,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.25+ },+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "audio", "video", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 16.7,+ "outputTokensPerSecond": 122.73,+ "costPerTask": 0.2317,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gemini-flash-latest",+ "label": "Gemini Flash Latest",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.75,+ "outputCostPerMillionTokens": 3.75,+ "cachedInputCostPerMillionTokens": 0.075,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "video", "audio", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 41.2,+ "outputTokensPerSecond": 261.3,+ "costPerTask": 1.2428,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "gemini-2.5-flash",+ "label": "Gemini 2.5 Flash",+ "modelFamily": "GEMINI",+ "inputCostPerMillionTokens": 0.3,+ "outputCostPerMillionTokens": 2.5,+ "cachedInputCostPerMillionTokens": 0.03,+ "contextWindowTokens": 1048576,+ "maxOutputTokens": 65536,+ "modalities": ["image", "audio", "video", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 13.1,+ "outputTokensPerSecond": 211.65,+ "measuredAt": "2026-09-13"+ }+ }+ ]+ },+ "mistral": {+ "models": [+ {+ "name": "pixtral-12b",+ "label": "Pixtral 12B",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.15,+ "outputCostPerMillionTokens": 0.15,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000,+ "modalities": ["image"]+ },+ {+ "name": "devstral-small-2507",+ "label": "Devstral Small",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.1,+ "outputCostPerMillionTokens": 0.3,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000,+ "isDeprecated": true+ },+ {+ "name": "mistral-small-2506",+ "label": "Mistral Small 3.2",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.1,+ "outputCostPerMillionTokens": 0.3,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "modalities": ["image"]+ },+ {+ "name": "magistral-small",+ "label": "Magistral Small",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.5,+ "outputCostPerMillionTokens": 1.5,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000,+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 8.2,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "devstral-2512",+ "label": "Devstral 2",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 2,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "isDeprecated": true+ },+ {+ "name": "devstral-small-2505",+ "label": "Devstral Small 2505",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.1,+ "outputCostPerMillionTokens": 0.3,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000,+ "benchmark": {+ "intelligenceIndex": 8.7,+ "measuredAt": "2026-09-13"+ },+ "isDeprecated": true+ },+ {+ "name": "labs-devstral-small-2512",+ "label": "Devstral Small 2",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0,+ "outputCostPerMillionTokens": 0,+ "contextWindowTokens": 256000,+ "maxOutputTokens": 256000,+ "modalities": ["image"],+ "isDeprecated": true+ },+ {+ "name": "magistral-medium-latest",+ "label": "Magistral Medium (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 5,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 16384,+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 9.1,+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "open-mixtral-8x22b",+ "label": "Mixtral 8x22B",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 6,+ "contextWindowTokens": 64000,+ "maxOutputTokens": 64000+ },+ {+ "name": "open-mixtral-8x7b",+ "label": "Mixtral 8x7B",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.7,+ "outputCostPerMillionTokens": 0.7,+ "contextWindowTokens": 32000,+ "maxOutputTokens": 32000+ },+ {+ "name": "open-mistral-7b",+ "label": "Mistral 7B",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.25,+ "outputCostPerMillionTokens": 0.25,+ "contextWindowTokens": 8000,+ "maxOutputTokens": 8000+ },+ {+ "name": "mistral-medium-latest",+ "label": "Mistral Medium (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 1.5,+ "outputCostPerMillionTokens": 7.5,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "devstral-medium-2507",+ "label": "Devstral Medium",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 2,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000,+ "isDeprecated": true+ },+ {+ "name": "mistral-medium-2604",+ "label": "Mistral Medium 3.5",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 1.5,+ "outputCostPerMillionTokens": 7.5,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "mistral-large-2512",+ "label": "Mistral Large 3",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.5,+ "outputCostPerMillionTokens": 1.5,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "modalities": ["image"]+ },+ {+ "name": "devstral-medium-latest",+ "label": "Devstral 2 (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 2,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "isDeprecated": true+ },+ {+ "name": "voxtral-small-latest",+ "label": "Voxtral Small (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.1,+ "outputCostPerMillionTokens": 0.3,+ "contextWindowTokens": 32000,+ "maxOutputTokens": 32000,+ "modalities": ["audio"]+ },+ {+ "name": "ministral-8b-latest",+ "label": "Ministral 8B (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.1,+ "outputCostPerMillionTokens": 0.1,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000+ },+ {+ "name": "mistral-nemo",+ "label": "Mistral Nemo",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.15,+ "outputCostPerMillionTokens": 0.15,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000+ },+ {+ "name": "mistral-small-latest",+ "label": "Mistral Small (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.15,+ "outputCostPerMillionTokens": 0.6,+ "contextWindowTokens": 256000,+ "maxOutputTokens": 256000,+ "modalities": ["image"],+ "supportsReasoning": true,+ "efforts": ["none", "high"]+ },+ {+ "name": "open-mistral-nemo",+ "label": "Open Mistral Nemo",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.15,+ "outputCostPerMillionTokens": 0.15,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000,+ "isDeprecated": true+ },+ {+ "name": "mistral-large-latest",+ "label": "Mistral Large (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.5,+ "outputCostPerMillionTokens": 1.5,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "modalities": ["image"]+ },+ {+ "name": "mistral-large-2411",+ "label": "Mistral Large 2.1",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 6,+ "contextWindowTokens": 131072,+ "maxOutputTokens": 16384+ },+ {+ "name": "devstral-latest",+ "label": "Devstral 2",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 2,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "isDeprecated": true+ },+ {+ "name": "mistral-small-2603",+ "label": "Mistral Small 4",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.15,+ "outputCostPerMillionTokens": 0.6,+ "contextWindowTokens": 256000,+ "maxOutputTokens": 256000,+ "modalities": ["image"],+ "supportsReasoning": true+ },+ {+ "name": "mistral-medium-2505",+ "label": "Mistral Medium 3",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 2,+ "contextWindowTokens": 131072,+ "maxOutputTokens": 131072,+ "modalities": ["image"]+ },+ {+ "name": "codestral-latest",+ "label": "Codestral (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.3,+ "outputCostPerMillionTokens": 0.9,+ "contextWindowTokens": 256000,+ "maxOutputTokens": 4096+ },+ {+ "name": "ministral-3b-latest",+ "label": "Ministral 3B (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.04,+ "outputCostPerMillionTokens": 0.04,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000+ },+ {+ "name": "mistral-medium-2508",+ "label": "Mistral Medium 3.1",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 0.4,+ "outputCostPerMillionTokens": 2,+ "contextWindowTokens": 262144,+ "maxOutputTokens": 262144,+ "modalities": ["image"]+ },+ {+ "name": "pixtral-large-latest",+ "label": "Pixtral Large (latest)",+ "modelFamily": "MISTRAL",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 6,+ "contextWindowTokens": 128000,+ "maxOutputTokens": 128000,+ "modalities": ["image"],+ "benchmark": {+ "intelligenceIndex": 7.1,+ "measuredAt": "2026-09-13"+ }+ }+ ]+ },+ "xai": {+ "models": [+ {+ "name": "grok-4.3",+ "label": "Grok 4.3",+ "modelFamily": "GROK",+ "inputCostPerMillionTokens": 1.25,+ "outputCostPerMillionTokens": 2.5,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 5,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 30000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "benchmark": {+ "intelligenceIndex": 25.4,+ "outputTokensPerSecond": 119.35,+ "costPerTask": 0.1653,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ },+ {+ "name": "grok-4.20-0309-reasoning",+ "label": "Grok 4.20 (Reasoning)",+ "modelFamily": "GROK",+ "inputCostPerMillionTokens": 1.25,+ "outputCostPerMillionTokens": 2.5,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 5,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 30000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "grok-4.5",+ "label": "Grok 4.5",+ "modelFamily": "GROK",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 6,+ "cachedInputCostPerMillionTokens": 0.3,+ "longContextCost": {+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 12,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.6+ },+ "contextWindowTokens": 500000,+ "maxOutputTokens": 500000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high"],+ "benchmark": {+ "intelligenceIndex": 39.1,+ "outputTokensPerSecond": 54.84,+ "costPerTask": 1.0365,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "high": {+ "intelligenceIndex": 39.1,+ "outputTokensPerSecond": 54.84,+ "costPerTask": 1.0365,+ "effort": "high",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "grok-build-0.1",+ "label": "Grok Build 0.1",+ "modelFamily": "GROK",+ "inputCostPerMillionTokens": 1,+ "outputCostPerMillionTokens": 2,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 4,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 256000,+ "maxOutputTokens": 256000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true+ },+ {+ "name": "grok-4.6",+ "label": "Grok 4.6",+ "modelFamily": "GROK",+ "inputCostPerMillionTokens": 2,+ "outputCostPerMillionTokens": 6,+ "cachedInputCostPerMillionTokens": 0.5,+ "longContextCost": {+ "inputCostPerMillionTokens": 4,+ "outputCostPerMillionTokens": 12,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 1+ },+ "contextWindowTokens": 500000,+ "maxOutputTokens": 500000,+ "modalities": ["image", "pdf"],+ "supportsReasoning": true,+ "efforts": ["low", "medium", "high", "xhigh"],+ "benchmark": {+ "intelligenceIndex": 44.4,+ "outputTokensPerSecond": 53.59,+ "costPerTask": 1.8589,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "benchmarkByEffort": {+ "low": {+ "intelligenceIndex": 35.4,+ "outputTokensPerSecond": 48.56,+ "costPerTask": 0.4753,+ "effort": "low",+ "measuredAt": "2026-09-13"+ },+ "medium": {+ "intelligenceIndex": 43,+ "outputTokensPerSecond": 49.99,+ "costPerTask": 1.4963,+ "effort": "medium",+ "measuredAt": "2026-09-13"+ },+ "high": {+ "intelligenceIndex": 44.4,+ "outputTokensPerSecond": 53.59,+ "costPerTask": 1.8589,+ "effort": "high",+ "measuredAt": "2026-09-13"+ },+ "xhigh": {+ "intelligenceIndex": 44.3,+ "outputTokensPerSecond": 54.05,+ "costPerTask": 2.3237,+ "effort": "xhigh",+ "measuredAt": "2026-09-13"+ }+ }+ },+ {+ "name": "grok-4.20-0309-non-reasoning",+ "label": "Grok 4.20 (Non-Reasoning)",+ "modelFamily": "GROK",+ "inputCostPerMillionTokens": 1.25,+ "outputCostPerMillionTokens": 2.5,+ "cachedInputCostPerMillionTokens": 0.2,+ "longContextCost": {+ "inputCostPerMillionTokens": 2.5,+ "outputCostPerMillionTokens": 5,+ "thresholdTokens": 200000,+ "cachedInputCostPerMillionTokens": 0.4+ },+ "contextWindowTokens": 1000000,+ "maxOutputTokens": 30000,+ "modalities": ["image", "pdf"],+ "benchmark": {+ "intelligenceIndex": 14.6,+ "effort": "none",+ "measuredAt": "2026-09-13"+ }+ }+ ]+ }+}diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-models/ai-self-host-spec.json b/packages/twenty-server/src/engine/metadata-modules/ai/ai-models/ai-self-host-spec.jsonnew file mode 100644index 00000000000..395a086e1f3--- /dev/null+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-models/ai-self-host-spec.json@@ -0,0 +1,49 @@+{+ "providers": [+ {+ "name": "openai",+ "npm": "@ai-sdk/openai",+ "label": "OpenAI",+ "apiKey": "{{OPENAI_API_KEY}}",+ "models": {+ "vendor": "openai"+ }+ },+ {+ "name": "anthropic",+ "npm": "@ai-sdk/anthropic",+ "label": "Anthropic",+ "apiKey": "{{ANTHROPIC_API_KEY}}",+ "models": {+ "vendor": "anthropic"+ }+ },+ {+ "name": "google",+ "npm": "@ai-sdk/google",+ "label": "Google",+ "apiKey": "{{GOOGLE_API_KEY}}",+ "models": {+ "vendor": "google"+ }+ },+ {+ "name": "mistral",+ "npm": "@ai-sdk/mistral",+ "label": "Mistral",+ "apiKey": "{{MISTRAL_API_KEY}}",+ "models": {+ "vendor": "mistral"+ }+ },+ {+ "name": "xai",+ "npm": "@ai-sdk/xai",+ "label": "xAI",+ "apiKey": "{{XAI_API_KEY}}",+ "models": {+ "vendor": "xai"+ }+ }+ ]+}