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.

fix(ci): restore automated LLM catalog updates (#10735)

Shahmir Varqhaac243eddc8f2
diff --git a/.github/workflows/sync-llm-info.yml b/.github/workflows/sync-llm-info.yml
index cd0bff0138..7bad242aae 100644
--- a/.github/workflows/sync-llm-info.yml
+++ b/.github/workflows/sync-llm-info.yml
@@ -77,7 +77,7 @@ jobs:
id: existing_pr
if: steps.changes.outputs.has_changes == 'true'
env:
- GH_TOKEN: ${{ steps.app-token.outputs.token }}
+ GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
run: |
number="$(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: ${{ steps.app-token.outputs.token }}
+ GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
run: |
gh pr create \
--draft \
diff --git a/packages/llm-info/data/models.yml b/packages/llm-info/data/models.yml
index 3cbe474f01..0ecbf6dc92 100644
--- a/packages/llm-info/data/models.yml
+++ b/packages/llm-info/data/models.yml
@@ -1,5 +1,15 @@
anthropic:
+ - name: Claude Fable 5.1
+ model: claude-fable-5-1
+ description: "Anthropic's most capable generally available model for ambitious, long-running coding, agentic, research, and enterprise workflows"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-09-01
+ cost: {input: 10, output: 50}
+
- name: Claude Opus 5
model: claude-opus-5
description: "Anthropic's flagship model for demanding reasoning, coding, and long-horizon agentic work"
@@ -12,7 +22,7 @@ anthropic:
- name: Claude Fable 5
model: claude-fable-5
- description: "Anthropic's most capable widely released Mythos-class model, built for the most demanding reasoning and long-horizon agentic work"
+ description: "Anthropic's Mythos-class model built for demanding reasoning, coding, and long-horizon agentic work"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -181,16 +191,6 @@ openai:
release_date: 2026-03-05
cost: {input: 2.5, output: 15}
- - name: GPT-5.3 Chat
- model: gpt-5.3-chat-latest
- description: "Update to ChatGPT's most-used model that makes everyday conversations smoother, more useful, and more directly helpful"
- roles: [chat, edit]
- capabilities: [tool_calling]
- input_types: [text, image]
- output_types: [text]
- release_date: 2026-03-03
- cost: {input: 1.75, output: 14}
-
- name: GPT-5.3 Codex
model: gpt-5.3-codex
description: "Agentic coding model that combines GPT-5.2-Codex software engineering performance with GPT-5.2 reasoning and professional knowledge"
@@ -211,6 +211,26 @@ openai:
release_date: 2025-12-16
google:
+ - name: Gemini 3.8 Flash
+ model: gemini-3.8-flash
+ description: "Google's most intelligent Flash model, with gains in software engineering, agentic tasks, and multi-step reasoning"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-09-02
+ cost: {input: 0.75, output: 3.75}
+
+ - name: Claude Fable 5.1
+ model: claude-fable-5-1@default
+ description: "Anthropic's most capable generally available model for ambitious, long-running coding, agentic, research, and enterprise workflows, available through Vertex AI only"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-09-01
+ cost: {input: 10, output: 50}
+
- 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"
@@ -360,6 +380,16 @@ google:
cost: {input: 5, output: 25}
bedrock:
+ - name: Claude Fable 5.1 (Global)
+ model: global.anthropic.claude-fable-5-1
+ description: "Anthropic's most capable generally available model for ambitious, long-running coding, agentic, research, and enterprise workflows"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-09-01
+ cost: {input: 10, output: 50}
+
- name: Grok 4.6
model: xai.grok-4.6
description: "xAI's frontier model for coding, agentic tasks, knowledge work, and visual reasoning"
@@ -432,7 +462,7 @@ bedrock:
- name: Claude Fable 5 (Global)
model: global.anthropic.claude-fable-5
- description: "Anthropic's most capable widely released Mythos-class model, built for the most demanding reasoning and long-horizon agentic work"
+ description: "Anthropic's Mythos-class model built for demanding reasoning, coding, and long-horizon agentic work"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -452,7 +482,7 @@ bedrock:
- name: Claude Fable 5 (Global)
model: global.anthropic.claude-fable-5
- description: "Anthropic's most capable widely released Mythos-class model, built for the most demanding reasoning and long-horizon agentic work"
+ description: "Anthropic's Mythos-class model built for demanding reasoning, coding, and long-horizon agentic work"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -541,6 +571,25 @@ bedrock:
cost: {input: 3, output: 15}
azure:
+ - name: Claude Fable 5.1
+ model: claude-fable-5-1
+ description: "Anthropic's most capable generally available model for ambitious, long-running coding, agentic, research, and enterprise workflows"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-09-01
+ cost: {input: 10, output: 50}
+
+ - name: Grok 4.6
+ model: grok-4.6
+ description: "xAI frontier model for coding, knowledge work, and STEM"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image]
+ output_types: [text]
+ release_date: 2026-08-12
+
- name: Claude Opus 5
model: claude-opus-5
description: "Anthropic's flagship model for demanding reasoning, coding, and long-horizon agentic work"
@@ -593,7 +642,7 @@ azure:
- name: Claude Fable 5
model: claude-fable-5
- description: "Anthropic's most capable widely released Mythos-class model, built for the most demanding reasoning and long-horizon agentic work"
+ description: "Anthropic's Mythos-class model built for demanding reasoning, coding, and long-horizon agentic work"
roles: [chat, edit]
capabilities: [thinking, tool_calling]
input_types: [text, image, pdf]
@@ -681,16 +730,6 @@ azure:
release_date: 2026-03-05
cost: {input: 2.5, output: 15}
- - name: GPT-5.3 Chat
- model: gpt-5.3-chat
- description: "Update to ChatGPT's most-used model that makes everyday conversations smoother, more useful, and more directly helpful"
- roles: [chat, edit]
- capabilities: [thinking, tool_calling]
- input_types: [text, image]
- output_types: [text]
- release_date: 2026-03-03
- cost: {input: 1.75, output: 14}
-
- name: GPT-5.3 Codex
model: gpt-5.3-codex
description: "Agentic coding model that combines GPT-5.2-Codex software engineering performance with GPT-5.2 reasoning and professional knowledge"
@@ -843,6 +882,24 @@ github:
cost: {input: 0, output: 0}
ollama:
+ - name: GLM-5.3-Flash
+ model: glm-5.3-flash
+ description: "Native multimodal model from Z.ai optimized for efficient coding and long-horizon agent tasks"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-08-26
+
+ - name: GLM-5.3
+ model: glm-5.3
+ description: "Large-scale reasoning model from Z.ai built for complex software engineering and long-horizon agent tasks"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text]
+ output_types: [text]
+ release_date: 2026-08-14
+
- name: DeepSeek V4 Flash 0731
model: deepseek-v4-flash:0731
description: "DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total"
@@ -970,6 +1027,26 @@ ollama:
release_date: 2026-02-15
wandb:
+ - name: Granite 4.2 8B
+ model: ibm-granite/granite-4.2-8b
+ description: "Dense reasoning model from IBM for mathematics, code generation, multilingual dialogue, and multi-step agentic workflows"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text]
+ output_types: [text]
+ release_date: 2026-08-24
+ cost: {input: 0.1, output: 0.15}
+
+ - name: DeepSeek V4 Pro 0813
+ model: deepseek-ai/DeepSeek-V4-Pro-0813
+ description: "Large-scale Mixture-of-Experts model and generally available release of DeepSeek V4 Pro"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text]
+ output_types: [text]
+ release_date: 2026-08-13
+ cost: {input: 1.31, output: 3.96}
+
- 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"
@@ -1161,25 +1238,55 @@ 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"
+ - name: Muse Spark 1.3 Contributor
+ model: muse-spark-1.3-contributor
+ description: "Cost-efficient contributor tier of Meta's multimodal reasoning model for experimentation, learning, and early-stage agentic, multi-agent, and coding workflows"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-09-02
+ cost: {input: 0.1, output: 0.2}
+
+ - name: Hy4 preview
+ model: hy4-preview
+ description: "Mixture-of-Experts model from Tencent designed for coding agents, complex tool use, and productivity tasks"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text]
+ output_types: [text]
+ release_date: 2026-08-28
+ cost: {input: 0.834, output: 2.501}
+
+ - name: GLM-5.3-Flash (2x usage)
+ model: glm-5.3-flash
+ description: "Native multimodal model from Z.ai optimized for efficient coding and long-horizon agent tasks"
+ roles: [chat, edit]
+ capabilities: [thinking, tool_calling]
+ input_types: [text, image, pdf]
+ output_types: [text]
+ release_date: 2026-08-26
+ cost: {input: 0.075, output: 0.25}
+
+ - name: Qwen3.8 Flash
+ model: qwen3.8-flash
+ description: "Multimodal reasoning model from Alibaba for coding, agentic workflows, visual understanding, document and codebase analysis, and desktop interaction"
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}
+ release_date: 2026-08-26
+ cost: {input: 0.15, output: 0.47}
- - name: Ox Alpha Free (Unlimited)
- model: ox-alpha-free
- description: "Free, limited-time OpenCode Go model for coding and agentic tasks"
+ - 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, output: 0}
+ cost: {input: 0.22, output: 0.66}
- name: Muse Spark 1.2 Contributor
model: muse-spark-1.2-contributor
@@ -1231,26 +1338,6 @@ opencode-go:
release_date: 2026-07-16
cost: {input: 3, output: 15}
- - name: Grok 4.5
- model: grok-4.5
- description: "xAI model for coding, knowledge work, and STEM"
- roles: [chat, edit]
- capabilities: [thinking, tool_calling]
- input_types: [text, image]
- output_types: [text]
- release_date: 2026-07-08
- cost: {input: 2, output: 6}
-
- - name: Hy3
- model: hy3
- description: "Tencent's 295B-parameter Mixture-of-Experts model (21B active) built for reasoning, agentic workflows, and production use"
- roles: [chat, edit]
- capabilities: [thinking, tool_calling]
- input_types: [text]
- output_types: [text]
- release_date: 2026-07-06
- cost: {input: 0.14, output: 0.58}
-
- name: DeepSeek V4 Flash
model: deepseek-v4-flash
description: "Efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window"
@@ -1301,16 +1388,6 @@ opencode-go:
release_date: 2026-06-13
cost: {input: 1.4, output: 4.4}
- - name: GLM-5.1
- model: glm-5.1
- description: "GLM-5.1 delivers a major leap in coding capability, with particularly significant gains in handling long-horizon tasks"
- roles: [chat, edit]
- capabilities: [thinking, tool_calling]
- input_types: [text]
- output_types: [text]
- release_date: 2026-04-07
- cost: {input: 1.4, output: 4.4}
-
- name: Qwen3.7 Plus
model: qwen3.7-plus
description: "Cost-effective model in Alibaba's Qwen3.7 series supporting text and image input with upgraded text capabilities"
@@ -1341,16 +1418,6 @@ opencode-go:
release_date: 2026-05-31
cost: {input: 0.1, output: 0.4}
- - name: MiniMax M2.7
- model: minimax-m2.7
- description: "Large language model designed for autonomous, real-world productivity and continuous improvement"
- roles: [chat, edit]
- capabilities: [thinking, tool_calling]
- input_types: [text]
- output_types: [text]
- release_date: 2026-03-18
- cost: {input: 0.3, output: 1.2}
-
- name: MiMo V2.5
model: mimo-v2.5
description: "Native omnimodal model from Xiaomi delivering Pro-level agentic performance at roughly half the inference cost"
diff --git a/packages/llm-info/skills/SKILL.md b/packages/llm-info/skills/SKILL.md
index c0c5cd1ae6..4911d5c624 100644
--- a/packages/llm-info/skills/SKILL.md
+++ b/packages/llm-info/skills/SKILL.md
@@ -1,6 +1,6 @@
---
name: fill-model-descriptions
-description: Fill empty `description: ""` fields in `packages/llm-info/data/models.yml` by querying OpenRouter and provider documentation. Use when the user asks to populate model descriptions, enrich the model catalog, or backfill descriptions after running `pnpm sync-models`.
+description: Fill missing and refresh obsolete model descriptions in `packages/llm-info/data/models.yml` by querying OpenRouter and provider documentation. Use when the user asks to populate model descriptions, enrich the model catalog, or curate descriptions after running `pnpm sync-models`.
disable-model-invocation: true
---
@@ -8,11 +8,13 @@ disable-model-invocation: true
## Goal
-Replace empty `description: ""` strings in `packages/llm-info/data/models.yml` with concise, accurate one-sentence summaries sourced from OpenRouter first, then provider documentation as a fallback.
+Keep descriptions in `packages/llm-info/data/models.yml` concise and accurate. Fill empty descriptions and refresh existing descriptions when newly added models or authoritative documentation make them obsolete. Source descriptions from OpenRouter first, then provider documentation as a fallback.
## Scope and Contract
-- **Only modify entries where `description` is the empty string.** Existing non-empty descriptions are human-curated; never overwrite them.
+- Fill every empty `description` for which a confident source is available.
+- Preserve human-curated, non-empty descriptions unless a newly added model or authoritative source makes a concrete claim obsolete. Typical obsolete claims include "latest," "most capable," availability, or capabilities that have demonstrably changed.
+- Do not rewrite existing descriptions only for style, tone, or consistency. When an existing description must change, make the smallest factual correction and record it in the final summary.
- The file is a top-level YAML map keyed by provider id (`anthropic:`, `openai:`, …). Each provider's value is an array of model entries.
- Do not modify any other field, reorder entries, or change formatting (flow-style arrays like `roles: [chat, edit]`, blank lines between entries, blank line between provider sections).
- This skill requires network access.
@@ -21,7 +23,9 @@ Replace empty `description: ""` strings in `packages/llm-info/data/models.yml` w
1. **Read** `packages/llm-info/data/models.yml`. Parse it with the `yaml` library's `parseDocument` (NOT `parse`) so formatting and comments are preserved on write-back.
-2. **Collect** every entry where `description` is the empty string. Track them as `{ provider, model }` pairs.
+2. **Collect and review** descriptions:
+ - Build one worklist of `{ provider, model }` pairs containing every entry where `description` is empty.
+ - Review related existing entries in the same model families for claims made obsolete by the new entries. Also review descriptions affected by a factual change found in authoritative documentation, and add every obsolete entry to the same worklist.
3. **Source descriptions** in this priority order:
@@ -46,7 +50,8 @@ Replace empty `description: ""` strings in `packages/llm-info/data/models.yml` w
The `schema.test.ts` test will catch any structural drift.
7. **Report** to the user:
- - How many entries got descriptions.
+ - How many empty descriptions were filled.
+ - How many obsolete descriptions were updated, with the provider/model list and the factual reason.
- How many were skipped (with the provider/model list).
## Vendor Map
@@ -78,8 +83,8 @@ OpenRouter id vendor prefixes for each marimo provider:
## Examples
**Good** (matches the tone of existing hand-curated entries):
-- `Latest Opus model, strongest for coding and long-running professional tasks`
-- `Most capable Sonnet-class model, with frontier performance across coding, agents, and professional work`
+- `Opus model optimized for coding and long-running professional tasks`
+- `Sonnet-class model with strong performance across coding, agents, and professional work`
- `Lightweight Gemma model trained by Google, designed to run on a single GPU`
- `Multimodal Mixture-of-Experts model optimized for complex tool use and reasoning`
diff --git a/packages/llm-info/src/__tests__/sync-models.test.ts b/packages/llm-info/src/__tests__/sync-models.test.ts
index 58d4b31111..8916b666b3 100644
--- a/packages/llm-info/src/__tests__/sync-models.test.ts
+++ b/packages/llm-info/src/__tests__/sync-models.test.ts
@@ -569,12 +569,17 @@ describe("mergeModels", () => {
expect(MODEL_DENYLIST["openai"]?.has("gpt-realtime-2.1")).toBe(true);
expect(MODEL_DENYLIST["bedrock"]).toEqual(
new Set([
+ "anthropic.claude-fable-5-1",
+ "us.anthropic.claude-fable-5-1",
"anthropic.claude-opus-5",
"au.anthropic.claude-opus-5",
"eu.anthropic.claude-opus-5",
"jp.anthropic.claude-opus-5",
]),
);
+ expect(MODEL_DENYLIST["opencode-go"]).toEqual(
+ new Set(["glm-5.1", "hy3", "minimax-m2.7"]),
+ );
});
it("drops upstream models that match the per-provider denylist", () => {
diff --git a/packages/llm-info/src/sources/merge.ts b/packages/llm-info/src/sources/merge.ts
index aeca0d1039..899fbc9bd2 100644
--- a/packages/llm-info/src/sources/merge.ts
+++ b/packages/llm-info/src/sources/merge.ts
@@ -51,13 +51,25 @@ export const MODEL_DENYLIST: Readonly<Record<string, ReadonlySet<string>>> = {
// Listed upstream but does not work with marimo's OpenAI integration.
"gpt-realtime-2.1",
]),
+ azure: new Set([
+ // This ChatGPT alias is not exposed in Azure's current model catalog.
+ "gpt-5.3-chat",
+ ]),
bedrock: new Set([
// Provider-specific aliases are redundant with the global catalog entry.
+ "anthropic.claude-fable-5-1",
+ "us.anthropic.claude-fable-5-1",
"anthropic.claude-opus-5",
"au.anthropic.claude-opus-5",
"eu.anthropic.claude-opus-5",
"jp.anthropic.claude-opus-5",
]),
+ "opencode-go": new Set([
+ // Older models intentionally omitted from marimo's curated catalog.
+ "glm-5.1",
+ "hy3",
+ "minimax-m2.7",
+ ]),
};
function isDenylisted(marimoProviderId: string, modelId: string): boolean {
@@ -201,8 +213,8 @@ export interface MergeOptions {
* Merge models.dev data into existing per-provider entries.
*
* Pipeline per provider:
- * 1. Collect upstream candidates, dropping anything above the cost ceilings
- * or explicitly listed in `MODEL_DENYLIST`.
+ * 1. Collect upstream candidates, dropping deprecated models, anything above
+ * the cost ceilings, or entries explicitly listed in `MODEL_DENYLIST`.
* 2. Sort newest-first and trim to `maxPerProvider` — this is what `-n N`
* means: "the N freshest models upstream is exposing right now."
* 3. Of that top-N, drop ids we already have locally; the remainder is what
@@ -252,6 +264,9 @@ export function mergeModels(
candidatesByProvider.get(marimoProviderId) ?? new Map<string, AiModel>();
for (const model of Object.values(provider.models)) {
+ if (model.status === "deprecated") {
+ continue;
+ }
if (!isWithinCostBudget(model)) {
continue;
}
diff --git a/packages/llm-info/src/sources/models-dev.ts b/packages/llm-info/src/sources/models-dev.ts
index d7f6e7691b..063a53e614 100644
--- a/packages/llm-info/src/sources/models-dev.ts
+++ b/packages/llm-info/src/sources/models-dev.ts
@@ -8,6 +8,7 @@ export const ModelsDevModelSchema = z.object({
name: z.string(),
reasoning: z.boolean().optional(),
tool_call: z.boolean().optional(),
+ status: z.string().optional(),
release_date: z.string().optional(),
modalities: z
.object({