OpenAI's Codex Security CLI and TypeScript SDK for finding, validating, and fixing security vulnerabilities. npm: https://www.npmjs.com/package/@openai/codex-security
TypeScript ai-securityapplication-securityclicode-scanningcodexcodex-securitycybersecuritydevsecopsnodejsnpmopenaisdksecuritytypescriptvulnerability-scanning
fix(sdk): use published Vertex AI prices for managed OpenAI models (#2363)
Michael D'Angelo37c1a5fd341a
diff --git a/sdk/typescript/docs/cli.md b/sdk/typescript/docs/cli.mdindex 9e16b0148..ae07272f6 100644--- a/sdk/typescript/docs/cli.md+++ b/sdk/typescript/docs/cli.md@@ -280,6 +280,45 @@ See the [Bedrock guide](https://github.com/openai/codex-security/blob/main/docs/for credential alternatives, restricted-model access, cost estimates, andreasoning-summary configuration.+### Google Cloud Vertex AI cost estimates++The SDK estimates Standard online inference costs for OpenAI models served by+Google Cloud's Model as a Service (MaaS). Rates are in USD per million tokens.+[Google Cloud partner model pricing](https://cloud.google.com/model-garden/partner-models/pricing)+and [OpenAI model locations](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/maas/openai).++| Managed model ID | Location | Input | Cached input | Output |+| -------------------------- | ----------------------- | ----: | -----------: | -----: |+| `openai/gpt-oss-120b-maas` | `global` | $0.09 | $0.09 | $0.36 |+| `openai/gpt-oss-20b-maas` | `us-central1` (retired) | $0.07 | $0.007 | $0.25 |++Google publishes a cache-hit discount only for 20B, so the estimator uses the+ordinary input rate for cached 120B tokens. Google retired the managed 20B+endpoint on October 21, 2026. The SDK retains its rates for historical usage.+[Google Cloud retirement schedule](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/open-models).++To estimate recorded usage locally, pass the managed model ID and the GCP+provider to `estimateScanCost`.++```ts+import { estimateScanCost } from "@openai/codex-security";++const cost = estimateScanCost(+ "openai/gpt-oss-120b-maas",+ { input_tokens: 1_000_000, output_tokens: 500_000 },+ { provider: "gcp" },+);+// cost?.estimatedUsd is 0.27.+```++Google documents the managed GPT OSS IDs on its shared chat-completions API.+The bundled Codex `gcp` provider uses Google's Responses API with a separate+frontier model catalog. Google has published no Standard token rates for that+catalog, including GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, and GPT-6 Luna.+Those scans record token usage and require a published price to use+`--max-cost`. Self-deployed Model Garden endpoints incur infrastructure charges.+[Google Cloud's preview announcement](https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-brings-openai-models-to-customers).+## Generate a security policyUse `policy` to draft or update `SECURITY.md` before a scan. The policy describes@@ -1143,6 +1182,7 @@ mean allowance was unavailable. There is no token-to-allowance conversion here.Cost displays use Standard token rates for the model. OpenAI models use[OpenAI API pricing](https://developers.openai.com/api/docs/pricing). OpenAImodels on Bedrock use [AWS pricing for the model, inference route, and region](https://github.com/openai/codex-security/blob/main/docs/bedrock.md#scan-cost-estimates).+For managed OpenAI models on Vertex AI, see [Google Cloud Vertex AI cost estimates](#google-cloud-vertex-ai-cost-estimates).The short/long-context range estimates usage at published rates. Other servicetiers, account discounts, and ChatGPT subscription allowance are outside thatestimate. Models without known prices have no estimate.diff --git a/sdk/typescript/scripts/package-dist-files.mjs b/sdk/typescript/scripts/package-dist-files.mjsindex 064f43def..a4387076b 100644--- a/sdk/typescript/scripts/package-dist-files.mjs+++ b/sdk/typescript/scripts/package-dist-files.mjs@@ -79,6 +79,7 @@ const distModules = ["cost-model","bedrock-pricing","bedrock-oss-pricing",+ "vertex-pricing","coverage-presentation","custom-validation","custom-validation-prompt",diff --git a/sdk/typescript/src/api.ts b/sdk/typescript/src/api.tsindex 6f4c62b2f..87541bbb6 100644--- a/sdk/typescript/src/api.ts+++ b/sdk/typescript/src/api.ts@@ -6282,7 +6282,11 @@ async function prepareScanOutputDir(}function turnPricingMetadata(context?: ScanPricingContext) {- return context === undefined ? {} : { bedrockRegion: context.region };+ if (context === undefined) return {};+ // Preserve the region field written on Bedrock scan results.+ return context.provider === "amazon-bedrock"+ ? { bedrockRegion: context.region }+ : { pricingContext: context };}function validateScanCostLimit(@@ -6298,8 +6302,16 @@ function validateScanCostLimit(pricingContext,) === null) {- const region = pricingContext?.region;- const location = region ? ` in ${region}` : "";+ const region =+ pricingContext?.provider === "amazon-bedrock"+ ? pricingContext.region+ : undefined;+ const location =+ pricingContext?.provider === "gcp"+ ? " on Google Cloud"+ : region+ ? ` in ${region}`+ : "";const detail =region === null? " Set the Bedrock billing region to check its regional price."diff --git a/sdk/typescript/src/config.ts b/sdk/typescript/src/config.tsindex 0f679535d..4bcba8873 100644--- a/sdk/typescript/src/config.ts+++ b/sdk/typescript/src/config.ts@@ -152,19 +152,47 @@ export function scanPricingContext(config: Readonly<JsonObject>,environment: Readonly<Record<string, string | undefined>>,): ScanPricingContext | undefined {- const selected = scanModelProvider(config);- if (selected !== "amazon-bedrock" && selected !== "amazon-bedrock-runtime") {- return undefined;- }+ const selected = scanModelProvider(config) ?? "openai";+ if (typeof selected !== "string") return undefined;const resolved = selectedScanProfile(config)? resolveCodexProfile(config): config;const providers = resolved["model_providers"];const provider = isObject(providers) ? providers[selected] : undefined;- return {- provider: "amazon-bedrock",- region: bedrockPricingRegion(provider, environment),- };+ if (selected === "amazon-bedrock" || selected === "amazon-bedrock-runtime") {+ return {+ provider: "amazon-bedrock",+ region: bedrockPricingRegion(provider, environment),+ };+ }+ // Native Codex only accepts inference routing for the built-in OpenAI+ // provider through the top-level openai_base_url setting.+ const baseUrl =+ selected === "openai"+ ? resolved["openai_base_url"]+ : isObject(provider)+ ? provider["base_url"]+ : undefined;+ if (baseUrl == null)+ return selected === "gcp" ? { provider: "gcp" } : undefined;++ const url = typeof baseUrl === "string" ? URL.parse(baseUrl) : null;+ const host = url?.hostname.replace(/\.$/, "") ?? "";+ if (host.endsWith(".prediction.vertexai.goog")) {+ return { provider: "gcp", managedEndpoint: false };+ }+ const vertexHost =+ url &&+ /^(?:[a-z0-9-]+-)?aiplatform(?:\.mtls)?\.googleapis\.com$/.test(host);+ if (selected !== "gcp" && !vertexHost) return undefined;+ // A dedicated endpoint bills the serving infrastructure. Only publisher and+ // shared OpenAPI endpoints identify the managed token-based service.+ const managedEndpoint =+ !!vertexHost &&+ /^\/v1(?:beta1)?\/projects\/[^/]+\/locations\/[^/]+\/(?:publishers\/openai\/v1|endpoints\/openapi)\/?$/.test(+ url.pathname,+ );+ return { provider: "gcp", managedEndpoint };}/** @internal Use the selected native provider's configured credential key. */diff --git a/sdk/typescript/src/cost-model.ts b/sdk/typescript/src/cost-model.tsindex 752e51c29..705141bcb 100644--- a/sdk/typescript/src/cost-model.ts+++ b/sdk/typescript/src/cost-model.ts@@ -1,11 +1,11 @@+import { vertexModelPricing } from "./vertex-pricing.js";import { bedrockModelPricing } from "./bedrock-pricing.js";import { isRecord } from "./record.js";import { isSafeNonNegativeInteger } from "./value.js";-export interface ScanPricingContext {- provider: "amazon-bedrock";- region: string | null;-}+export type ScanPricingContext =+ | { provider: "amazon-bedrock"; region: string | null }+ | { provider: "gcp"; managedEndpoint?: boolean };export interface ScanCost {model: string;@@ -143,6 +143,22 @@ export function tokenUsage(value: unknown): ScanTokenUsage | null {};}+function providerModelPricing(+ model: string,+ context?: ScanPricingContext | string | null,+) {+ if (context == null || typeof context === "string") {+ return bedrockModelPricing(model, context);+ }+ if (context.provider === "amazon-bedrock") {+ return bedrockModelPricing(model, context.region);+ }+ if (context.provider === "gcp" && context.managedEndpoint !== false) {+ return vertexModelPricing(model);+ }+ return undefined;+}+/*** Estimates Standard token costs using the selected provider's published rates.* A string (or null for an unknown region) selects Bedrock for compatibility.@@ -154,9 +170,7 @@ export function estimateScanCost(context?: ScanPricingContext | string | null,): ScanCost | null {if (model === undefined) return null;- const region =- typeof context === "object" && context !== null ? context.region : context;- const providerPricing = bedrockModelPricing(model, region);+ const providerPricing = providerModelPricing(model, context);const unitsPerUsd = providerPricing?.unitsPerUsd ?? 1_000_000_000;const pricing =providerPricing?.short ??diff --git a/sdk/typescript/src/result.ts b/sdk/typescript/src/result.tsindex 31fca3fe6..957e7b0fb 100644--- a/sdk/typescript/src/result.ts+++ b/sdk/typescript/src/result.ts@@ -1,3 +1,4 @@+import type { ScanPricingContext } from "./cost-model.js";import { statSync } from "node:fs";import { join } from "node:path";import type {@@ -15,6 +16,7 @@ export interface TurnResultMetadata {id?: string;status?: string;model?: string;+ pricingContext?: ScanPricingContext;/** Source region recorded for Bedrock scans. */bedrockRegion?: string | null;durationMs?: number;@@ -93,7 +95,7 @@ export class ScanResult {estimateScanCost(options.turnResult.model,options.turnResult.usage,- options.turnResult.bedrockRegion,+ options.turnResult.pricingContext ?? options.turnResult.bedrockRegion,);if (options.sarifPath !== undefined) {this.sarifPath = options.sarifPath;diff --git a/sdk/typescript/src/vertex-pricing.ts b/sdk/typescript/src/vertex-pricing.tsnew file mode 100644index 000000000..df1ea6cf3--- /dev/null+++ b/sdk/typescript/src/vertex-pricing.ts@@ -0,0 +1,27 @@+// Standard MaaS prices in nanodollars per token. Google lists one price+// across each model's 131,072-token context window. Only 20B discounts+// implicit cache hits. Keep its rates for scans made before its retirement.+const rates = new Map<+ string,+ readonly [+ input: number,+ cacheRead: number,+ cacheWrite: number,+ output: number,+ ]+>([+ ["gpt-oss-20b-maas", [70, 7, 70, 250]],+ ["gpt-oss-120b-maas", [90, 90, 90, 360]],+]);++export function vertexModelPricing(model: string) {+ const price = rates.get(model.replace(/^openai\//, ""));+ if (!price) return undefined;+ return {+ short: price,+ long: price,+ unitsPerUsd: 1_000_000_000,+ source: "https://cloud.google.com/model-garden/partner-models/pricing",+ asOf: "2026-11-05",+ };+}diff --git a/sdk/typescript/tests-ts/api.test.ts b/sdk/typescript/tests-ts/api.test.tsindex e9d763f1a..4d86cad7f 100644--- a/sdk/typescript/tests-ts/api.test.ts+++ b/sdk/typescript/tests-ts/api.test.ts@@ -14470,13 +14470,20 @@ if __name__ == "__main__":}});- test("reports unpriced model usage without letting its observer stop the scan", async () => {+ test("reports unpriced GCP usage without letting its observer stop the scan", async () => {const { root, repository, codexHome, scanDir } = await scanDirectories();const usages: unknown[] = [];const observerErrors: string[] = [];const commands: Array<readonly string[]> = [];const client = new TestClient(- { codexOverrides: { model: "synthetic-unpriced-model" } },+ {+ codexOverrides: {+ model: "gpt-6-astra",+ model_provider: "openai",+ profile: "review",+ profiles: { review: { model_provider: "gcp" } },+ },+ },{...scanRuntimeDependencies(codexHome, scanDir),runWorkbench: recordingWorkbench(commands),@@ -14484,6 +14491,13 @@ if __name__ == "__main__":},);try {+ await expect(+ client.preflight(repository, { maxCostUsd: 1 }),+ ).rejects.toThrow("cost limit is not available for the configured model");+ await expect(client.preflight(repository)).resolves.toMatchObject({+ model: "gpt-6-astra",+ modelProvider: "gcp",+ });const result = await client.run(repository, {onUsage: async (usage) => {usages.push(usage);@@ -14516,9 +14530,15 @@ if __name__ == "__main__":}});const prices = [- { model: "gpt-6-astra", min: 0.000232, max: 0.000429 },- { model: "gpt-6-cyber", min: 0.000464, max: null },- { model: "gpt-5.6-cyber", min: 0.0003275, max: null },+ { model: "gpt-6-astra", provider: "openai", min: 0.000232, max: 0.000429 },+ { model: "gpt-6-cyber", provider: "openai", min: 0.000464, max: null },+ { model: "gpt-5.6-cyber", provider: "openai", min: 0.0003275, max: null },+ {+ model: "openai/gpt-oss-120b-maas",+ provider: "gcp",+ min: 0.00000198,+ max: 0.00000198,+ },];test.each(prices)("tracks live and saved $model costs", async (price) => {const { model, min, max } = price;@@ -14535,6 +14555,7 @@ if __name__ == "__main__":profiles: {review: {model,+ model_provider: price.provider,model_reasoning_effort: "high",},},diff --git a/sdk/typescript/tests-ts/config.test.ts b/sdk/typescript/tests-ts/config.test.tsindex a6558936c..2670f2509 100644--- a/sdk/typescript/tests-ts/config.test.ts+++ b/sdk/typescript/tests-ts/config.test.ts@@ -706,6 +706,65 @@ describe("Codex configuration", () => {},);+ test("distinguishes managed Vertex APIs from dedicated endpoints and other gateways", () => {+ const base =+ "https://us-central1-aiplatform.googleapis.com/v1beta1/projects/example/locations/us-central1";+ const fromUrl = (baseUrl: string) =>+ scanPricingContext({ openai_base_url: baseUrl }, {});+ expect(fromUrl(`${base}/endpoints/openapi`)).toEqual({+ provider: "gcp",+ managedEndpoint: true,+ });+ expect(fromUrl(`${base}/endpoints/1234`)).toEqual({+ provider: "gcp",+ managedEndpoint: false,+ });+ expect(+ fromUrl("https://123.us-central1-example.prediction.vertexai.goog"),+ ).toEqual({ provider: "gcp", managedEndpoint: false });+ expect(+ fromUrl(+ "https://aiplatform.mtls.googleapis.com./v1/projects/example/locations/global/endpoints/openapi",+ ),+ ).toEqual({ provider: "gcp", managedEndpoint: true });+ expect(fromUrl("https://gateway.example.test")).toBeUndefined();+ expect(+ scanPricingContext(+ { model_providers: { openai: { base_url: base } } },+ { OPENAI_BASE_URL: base },+ ),+ ).toBeUndefined();+ const config = {+ profile: "review",+ profiles: {+ review: {+ model_provider: "vertex",+ model_providers: {+ vertex: {+ base_url:+ "https://aiplatform.googleapis.com/v1/projects/example/locations/global/publishers/openai/v1/",+ },+ },+ },+ },+ };+ expect(scanPricingContext(config, {})).toEqual({+ provider: "gcp",+ managedEndpoint: true,+ });+ expect(+ scanPricingContext(+ {+ model_provider: "gcp",+ model_providers: {+ gcp: { base_url: "https://gateway.example.test" },+ },+ },+ {},+ ),+ ).toEqual({ provider: "gcp", managedEndpoint: false });+ });+test("rejects invalid model settings from the selected Codex profile", async () => {for (const [profile, message] of [[{ model: " " }, "model must be a nonempty string"],diff --git a/sdk/typescript/tests-ts/scan-dashboard.test.ts b/sdk/typescript/tests-ts/scan-dashboard.test.tsindex b549b8f8e..8587dd858 100644--- a/sdk/typescript/tests-ts/scan-dashboard.test.ts+++ b/sdk/typescript/tests-ts/scan-dashboard.test.ts@@ -744,6 +744,8 @@ describe("live scan dashboard", () => {{ provider: "amazon-bedrock", region: "us-gov-west-1" },false,],+ ["gpt-6-astra", { provider: "gcp" }, false],+ ["openai/gpt-oss-120b-maas", { provider: "gcp" }, true],] as const)("shows a cost row only when %s has pricing in %j",(model, pricingContext, priced) => {diff --git a/sdk/typescript/tests-ts/vertex-pricing.test.ts b/sdk/typescript/tests-ts/vertex-pricing.test.tsnew file mode 100644index 000000000..d6750b783--- /dev/null+++ b/sdk/typescript/tests-ts/vertex-pricing.test.ts@@ -0,0 +1,56 @@+import { expect, test } from "bun:test";+import { estimateScanCost } from "../src/cost-model.js";++const usage = {+ input_tokens: 16_000,+ cached_input_tokens: 8_000,+ cache_write_input_tokens: 2_000,+ output_tokens: 1_000,+};++test.each([+ ["gpt-oss-20b-maas", 0.07, 0.007, 0.25, 0.000866],+ ["gpt-oss-120b-maas", 0.09, 0.09, 0.36, 0.0018],+] as const)(+ "estimates Vertex Standard usage for %s",+ (id, input, cacheRead, output, cost) => {+ for (const model of [id, `openai/${id}`]) {+ const estimate = estimateScanCost(model, usage, { provider: "gcp" });+ expect(estimate).toMatchObject({+ model,+ estimatedUsd: cost,+ estimatedUsdRange: { min: cost, max: cost },+ pricing: {+ source:+ "https://cloud.google.com/model-garden/partner-models/pricing",+ serviceTier: "standard",+ usdPerMillionTokens: { input, cacheRead, cacheWrite: input, output },+ },+ });+ }+ },+);++test.each([+ "gpt-oss-20b",+ "openai/gpt-oss-120b",+ "gpt-oss-20b-maas:latest",+ "meta/gpt-oss-20b-maas",+ "gpt-6-sol",+ "openai.gpt-5.6-cyber",+])("leaves GCP usage for %s unpriced", (model) => {+ expect(estimateScanCost(model, usage, { provider: "gcp" })).toBeNull();+});++test("requires a managed Vertex billing context for MaaS prices", () => {+ expect(estimateScanCost("openai/gpt-oss-120b-maas", usage)).toBeNull();+ expect(+ estimateScanCost("openai/gpt-oss-120b-maas", usage, {+ provider: "gcp",+ managedEndpoint: false,+ }),+ ).toBeNull();+ expect(+ estimateScanCost("openai/gpt-oss-20b-maas", usage, "us-east-1"),+ ).toBeNull();+});