Skip to content

A reactive notebook for Python — run reproducible experiments, query with SQL, execute as a script, deploy as an app, and version with git. Stored as pure Python. All in a modern, AI-native editor.

Update LLM model catalog (#10634)

github-actions[bot]5efd56bd598d
diff --git a/.github/workflows/sync-llm-info.yml b/.github/workflows/sync-llm-info.yml
index d7841e9254..cd0bff0138 100644
--- a/.github/workflows/sync-llm-info.yml
+++ b/.github/workflows/sync-llm-info.yml
@@ -53,15 +53,15 @@ jobs:
- name: 📥 Install repo
uses: ./.github/actions/install
- - name: 🔄 Sync models (5 newest per provider)
- run: pnpm --dir packages/llm-info sync-models -n 5
-
- name: 🛠️ Generate llm-info data
run: pnpm --dir packages/llm-info codegen
- name: 🧪 Test llm-info
run: pnpm --dir packages/llm-info test
+ - name: 🔄 Sync models (5 newest per provider)
+ run: pnpm --dir packages/llm-info sync-models -n 5
+
- name: 🔎 Check for model catalog changes
id: changes
run: |
@@ -77,7 +77,7 @@ jobs:
id: existing_pr
if: steps.changes.outputs.has_changes == 'true'
env:
- GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+ GH_TOKEN: ${{ steps.app-token.outputs.token }}
run: |
number="$(gh pr list --state open --head "$BRANCH_NAME" --json number --jq '.[0].number // ""')"
echo "number=$number" >> "$GITHUB_OUTPUT"
@@ -102,7 +102,7 @@ jobs:
- name: 📝 Create update PR
if: steps.changes.outputs.has_changes == 'true' && steps.existing_pr.outputs.number == ''
env:
- GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+ GH_TOKEN: ${{ steps.app-token.outputs.token }}
run: |
gh pr create \
--draft \
@@ -117,6 +117,12 @@ jobs:
- Syncs `packages/llm-info/data/models.yml` from `models.dev`.
- Runs llm-info code generation and tests.
+ ## Human review required
+ - Replace every empty model description with clear, user-facing prose.
+ - Review related models for descriptions that became outdated after this sync.
+ - Prefer concrete capabilities and intended uses.
+ - Remove redundant aliases and models that do not belong in marimo's catalog.
+
## Test plan
- `pnpm --dir packages/llm-info test`
EOF
diff --git a/.github/workflows/test_fe.yaml b/.github/workflows/test_fe.yaml
index 5b7310c8ba..b26b02b350 100644
--- a/.github/workflows/test_fe.yaml
+++ b/.github/workflows/test_fe.yaml
@@ -171,6 +171,10 @@ jobs:
TURBO_TOKEN: ${{ steps.secrets.outputs.TURBO_TOKEN }}
run: pnpm turbo test:coverage
+ - name: 🧪 Test llm-info
+ working-directory: ${{ github.workspace }}
+ run: pnpm --dir packages/llm-info test -- --run
+
- name: 📊 Report coverage
# Runs even when tests fail (reportOnFailure writes the summary anyway).
# Fork PRs run with a read-only GITHUB_TOKEN, so posting the comment
diff --git a/packages/llm-info/data/models.yml b/packages/llm-info/data/models.yml
index b1a401b89e..0e0b5b905b 100644
--- a/packages/llm-info/data/models.yml
+++ b/packages/llm-info/data/models.yml
@@ -42,7 +42,7 @@ anthropic:
- name: Claude Opus 4.7
model: claude-opus-4-7
- description: "Next generation of Anthropic's Opus family, built for long-running, asynchronous agents"
+ description: "Anthropic Opus model built for long-running, asynchronous agents"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -52,7 +52,7 @@ anthropic:
- name: Claude Sonnet 4.6
model: claude-sonnet-4-6
- description: "Anthropic's most capable Sonnet-class model yet, with frontier performance across coding, agents, and professional work"
+ description: "Anthropic Sonnet model for coding, agents, and professional work"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -62,7 +62,7 @@ anthropic:
- name: Claude Opus 4.6
model: claude-opus-4-6
- description: "Anthropic's strongest model for coding and long-running professional tasks"
+ description: "Anthropic Opus model for coding and long-running professional tasks"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -72,7 +72,7 @@ anthropic:
- name: Claude Opus 4.5
model: claude-opus-4-5
- description: "Anthropic's frontier reasoning model optimized for complex software engineering, agentic workflows, and long-horizon computer use"
+ description: "Anthropic Opus reasoning model for complex software engineering, agentic workflows, and long-horizon computer use"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -92,7 +92,7 @@ anthropic:
- name: Claude Sonnet 4.5
model: claude-sonnet-4-5
- description: "Anthropic's most advanced Sonnet model to date, optimized for real-world agents and coding workflows"
+ description: "Anthropic Sonnet model optimized for real-world agents and coding workflows"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -143,7 +143,7 @@ openai:
- name: GPT-5.5
model: gpt-5.5
- description: "OpenAI's frontier model designed for complex professional workloads, building on GPT-5.4 with stronger reasoning, higher reliability, and improved token efficiency on hard tasks"
+ description: "OpenAI model for complex professional workloads, with stronger reasoning, reliability, and token efficiency than GPT-5.4 on hard tasks"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -173,7 +173,7 @@ openai:
- name: GPT-5.4
model: gpt-5.4
- description: "OpenAI's latest frontier model, unifying the Codex and GPT lines into a single system"
+ description: "OpenAI model that unified the Codex and GPT lines into a single system"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -193,7 +193,7 @@ openai:
- name: GPT-5.3 Codex
model: gpt-5.3-codex
- description: "OpenAI's most advanced agentic coding model, combining the frontier software engineering performance of GPT-5.2-Codex with the broader reasoning and professional knowledge capabilities of GPT-5.2"
+ description: "Agentic coding model that combines GPT-5.2-Codex software engineering performance with GPT-5.2 reasoning and professional knowledge"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -211,6 +211,16 @@ openai:
release_date: 2025-12-16
google:
+ - name: Gemini Flash Latest
+ model: gemini-flash-latest
+ description: "Alias that points to the latest Gemini Flash release and can change when Google updates the model family"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-08-13
+ cost: {input: 0.75, output: 3.75}
+
- name: Gemini 3.7 Flash
model: gemini-3.7-flash
description: "High-efficiency Gemini model for agentic workflows, coding, and multimodal reasoning"
@@ -350,6 +360,36 @@ google:
cost: {input: 5, output: 25}
bedrock:
+ - name: Grok 4.6
+ model: xai.grok-4.6
+ description: "xAI's frontier model for coding, agentic tasks, knowledge work, and visual reasoning"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image]
+ output_types: [text]
+ release_date: 2026-08-12
+ cost: {input: 2.2, output: 6.6}
+
+ - name: GPT-5.6 Luna (Global)
+ model: global.openai.gpt-5.6-luna
+ description: "Fast, cost-efficient model in OpenAI's GPT-5.6 series, suited for high-volume, latency-sensitive chat and lightweight agentic workflows"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-07-09
+ cost: {input: 0.22, output: 1.32}
+
+ - name: GPT-5.6 Sol (Global)
+ model: global.openai.gpt-5.6-sol
+ description: "Flagship model in OpenAI's GPT-5.6 series, suited for complex reasoning, coding, and agentic workflows"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-07-09
+ cost: {input: 5.5, output: 33}
+
- name: Claude Opus 5 (Global)
model: global.anthropic.claude-opus-5
description: "Anthropic's flagship model for demanding reasoning, coding, and long-horizon agentic work"
@@ -442,7 +482,7 @@ bedrock:
- name: Claude Opus 4.7
model: anthropic.claude-opus-4-7
- description: "Next generation of Anthropic's Opus family, built for long-running, asynchronous agents"
+ description: "Anthropic Opus model built for long-running, asynchronous agents"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -452,7 +492,7 @@ bedrock:
- name: Claude Opus 4.7 (Global)
model: global.anthropic.claude-opus-4-7
- description: "Next generation of Anthropic's Opus family, built for long-running, asynchronous agents"
+ description: "Anthropic Opus model built for long-running, asynchronous agents"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -462,7 +502,7 @@ bedrock:
- name: MiniMax M2.5
model: minimax.minimax-m2.5
- description: "SOTA large language model designed for real-world productivity"
+ description: "Large language model designed for real-world productivity"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text]
@@ -492,7 +532,7 @@ bedrock:
- name: Claude Sonnet 4.6
model: anthropic.claude-sonnet-4-6
- description: "Anthropic's most capable Sonnet-class model yet, with frontier performance across coding, agents, and professional work"
+ description: "Anthropic Sonnet model for coding, agents, and professional work"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -573,7 +613,7 @@ azure:
- name: GPT-5.5
model: gpt-5.5
- description: "OpenAI's frontier model designed for complex professional workloads, building on GPT-5.4 with stronger reasoning, higher reliability, and improved token efficiency on hard tasks"
+ description: "OpenAI model for complex professional workloads, with stronger reasoning, reliability, and token efficiency than GPT-5.4 on hard tasks"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -583,7 +623,7 @@ azure:
- name: Kimi K2.6
model: kimi-k2.6
- description: "Moonshot AI's next-generation multimodal model, designed for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration"
+ description: "Moonshot AI multimodal model for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image]
@@ -633,7 +673,7 @@ azure:
- name: GPT-5.4
model: gpt-5.4
- description: "OpenAI's latest frontier model, unifying the Codex and GPT lines into a single system"
+ description: "OpenAI model that unified the Codex and GPT lines into a single system"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -653,7 +693,7 @@ azure:
- name: GPT-5.3 Codex
model: gpt-5.3-codex
- description: "OpenAI's most advanced agentic coding model, combining the frontier software engineering performance of GPT-5.2-Codex with the broader reasoning and professional knowledge capabilities of GPT-5.2"
+ description: "Agentic coding model that combines GPT-5.2-Codex software engineering performance with GPT-5.2 reasoning and professional knowledge"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image]
@@ -673,7 +713,7 @@ azure:
- name: Claude Sonnet 4.6
model: claude-sonnet-4-6
- description: "Anthropic's most capable Sonnet-class model yet, with frontier performance across coding, agents, and professional work"
+ description: "Anthropic Sonnet model for coding, agents, and professional work"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -714,7 +754,7 @@ github:
- name: Mistral Medium 3 (25.05)
model: mistral-ai/mistral-medium-2505
- description: "High-performance enterprise-grade language model designed to deliver frontier-level capabilities at significantly reduced operational cost"
+ description: "Enterprise language model designed for strong performance at reduced operational cost"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image]
@@ -754,7 +794,7 @@ github:
- name: DeepSeek-V3-0324
model: deepseek/deepseek-v3-0324
- description: "Latest model from the DeepSeek team, building upon the instruction following and coding abilities of the previous versions"
+ description: "DeepSeek V3 update with improved instruction following and coding capabilities"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text]
@@ -850,7 +890,7 @@ ollama:
- name: Kimi K2.6
model: kimi-k2.6
- description: "Moonshot AI's next-generation multimodal model, designed for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration"
+ description: "Moonshot AI multimodal model for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image]
@@ -895,7 +935,7 @@ ollama:
- name: MiniMax M2.7
model: minimax-m2.7
- description: "Next-generation large language model designed for autonomous, real-world productivity and continuous improvement"
+ description: "Large language model designed for autonomous, real-world productivity and continuous improvement"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text]
@@ -904,7 +944,7 @@ ollama:
- name: NVIDIA Nemotron 3 Ultra
model: nemotron-3-ultra
- description: "Open frontier-reasoning and orchestration model from NVIDIA with 55B active parameters out of 550B total (MoE)"
+ description: "Open reasoning and orchestration model from NVIDIA with 55B active parameters out of 550B total (MoE)"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text]
@@ -930,6 +970,16 @@ ollama:
release_date: 2026-02-15
wandb:
+ - name: Qwen3.8 27B
+ model: Qwen/Qwen3.8-27B
+ description: "Alibaba's 27B dense multimodal model for coding, professional work, research, and long-horizon agentic tasks"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image]
+ output_types: [text]
+ release_date: 2026-08-14
+ cost: {input: 0.4, output: 3}
+
- name: Nemotron 3.5 Lightning
model: nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B
description: "MoE model built for fast, reliable agentic tasks across financial services, cybersecurity, telecom, and retail"
@@ -1022,7 +1072,7 @@ wandb:
- name: Nemotron 3 Ultra
model: nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B
- description: "Open frontier-reasoning and orchestration model from NVIDIA with 55B active parameters out of 550B total (MoE)"
+ description: "Open reasoning and orchestration model from NVIDIA with 55B active parameters out of 550B total (MoE)"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text]
@@ -1042,7 +1092,7 @@ wandb:
- name: MiniMax M2.5
model: MiniMaxAI/MiniMax-M2.5
- description: "SOTA large language model designed for real-world productivity"
+ description: "Large language model designed for real-world productivity"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text]
@@ -1121,6 +1171,36 @@ wandb:
cost: {input: 0.05, output: 0.1}
opencode-go:
+ - name: DeepSeek V4 Flash Vision Exp
+ model: deepseek-v4-flash-vision-exp
+ description: "Experimental multimodal variant of DeepSeek V4 Flash that adds image understanding while retaining its text and agent capabilities"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image]
+ output_types: [text]
+ release_date: 2026-08-21
+ cost: {input: 0.22, output: 0.66}
+
+ - name: Ox Alpha Free (Unlimited)
+ model: ox-alpha-free
+ description: "Free, limited-time OpenCode Go model for coding and agentic tasks"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image]
+ output_types: [text]
+ release_date: 2026-08-21
+ cost: {input: 0, output: 0}
+
+ - name: Muse Spark 1.2 Contributor
+ model: muse-spark-1.2-contributor
+ description: "Discounted Meta coding model that permits prompts and completions to train future Meta models"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-08-05
+ cost: {input: 0.1, output: 0.2}
+
- name: GLM-5.3
model: glm-5.3
description: "Flagship GLM model for long-horizon coding, agents, and complex project delivery"
@@ -1163,7 +1243,7 @@ opencode-go:
- name: Grok 4.5
model: grok-4.5
- description: "xAI's smartest model with frontier performance on coding, knowledge work, and STEM"
+ description: "xAI model for coding, knowledge work, and STEM"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image]
@@ -1213,7 +1293,7 @@ opencode-go:
- name: Kimi K2.6
model: kimi-k2.6
- description: "Moonshot AI's next-generation multimodal model, designed for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration"
+ description: "Moonshot AI multimodal model for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image]
@@ -1273,7 +1353,7 @@ opencode-go:
- name: MiniMax M2.7
model: minimax-m2.7
- description: "Next-generation large language model designed for autonomous, real-world productivity and continuous improvement"
+ description: "Large language model designed for autonomous, real-world productivity and continuous improvement"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text]
diff --git a/packages/llm-info/src/__tests__/schema.test.ts b/packages/llm-info/src/__tests__/schema.test.ts
index 45792d1c05..1f648d75d3 100644
--- a/packages/llm-info/src/__tests__/schema.test.ts
+++ b/packages/llm-info/src/__tests__/schema.test.ts
@@ -96,4 +96,21 @@ describe("Data Validation", () => {
expect(Array.isArray(data)).toBe(false);
expect(() => ModelsByProviderSchema.parse(data)).not.toThrow();
});
+
+ it("requires a non-empty description for every model", () => {
+ const modelsPath = join(__dirname, "../../data/models.yml");
+ const modelsYaml = readFileSync(modelsPath, "utf-8");
+ const data = ModelsByProviderSchema.parse(parse(modelsYaml));
+ const modelsWithoutDescriptions = Object.entries(data).flatMap(
+ ([provider, models]) =>
+ models
+ .filter((model) => model.description.trim().length === 0)
+ .map((model) => `${provider}/${model.model}`),
+ );
+
+ expect(
+ modelsWithoutDescriptions,
+ "Every model must have a non-empty description",
+ ).toEqual([]);
+ });
});