models-dev-query

name	models-dev-query
description	Query AI model specifications from the models.dev database. Use when the user asks about model specs (pricing, context limits, capabilities: reasoning, tool_call, modalities), needs API endpoint information for a provider, wants to list what models a provider supports, or compare models across providers. Triggers for questions like "查一下 XXX 支持哪些模型"、"gpt-5 的定价"、 "配置 Trae 用 deepseek 的配置"。

Models.dev Query

Query models.dev — a comprehensive database of AI model specs. All data is fetched live from GitHub API or the aggregated JSON endpoint; no local data required.

Data Sources

Source	Command	Best For
Aggregated JSON API	curl -sL https://models.dev/api.json | jq '...'	Fast queries: list providers, get pricing/limits/capabilities
GitHub API (TOML files)	gh api repos/anomalyco/models.dev/contents/...	Detailed specs: reasoning mode, interleaved field, modalities, family

Default branch is dev (the GitHub repo's default). The main branch does not exist.

Quick Reference Commands

List all providers

curl -sL https://models.dev/api.json | jq -r 'keys | .[]' | sort

List models for a provider

# From api.json (fast)
curl -sL https://models.dev/api.json | jq -r '.["<provider-id>"].models | keys | .[]'

# From GitHub (detailed)
gh api repos/anomalyco/models.dev/contents/providers/<provider-id>/models \
  --jq '.[].name' | sed 's/\.toml$//'

Get model specs (fast — api.json)

curl -sL https://models.dev/api.json | jq '.["<provider-id>"].models["<model-id>"]'

Get model specs (detailed — GitHub TOML)

gh api repos/anomalyco/models.dev/contents/providers/<provider-id>/models/<model-id>.toml \
  --jq '.download_url' | xargs curl -sL

Get provider info

curl -sL https://models.dev/api.json | jq '.["<provider-id>"] | {name: .name, api: .api, doc: .doc, env: .env}'

Query Guide

1. "查一下 X 提供商支持哪些模型"

Use api.json to get the provider's model list + basic info in one shot:

curl -sL https://models.dev/api.json | jq -r '
  .["<provider-id>"] as $p |
  "Provider: \($p.name // "<provider-id>")",
  "API: \($p.api // "N/A")",
  "Models:",
  ($p.models | to_entries[] | "  - \(.key): \(.value.name)  (ctx: \(.value.limit.context // "?"), $\(.value.cost.input)/M in, $\(.value.cost.output)/M out)")
'

Format the output as a readable table.

2. "查某模型的详细参数"

Always fetch both sources for completeness:

# 1. Fast check from api.json
curl -sL https://models.dev/api.json | jq '.["<provider-id>"].models["<model-id>"]'

# 2. Detailed TOML for fields api.json doesn't have (family, reasoning, modalities, family, etc.)
gh api repos/anomalyco/models.dev/contents/providers/<provider-id>/models/<model-id>.toml \
  --jq '.download_url' | xargs curl -sL

If the model ID contains / (e.g. nvidia/llama-3.1), the TOML file is in a subdirectory:

providers/<provider-id>/models/<slash-prefix>/<rest-of-id>.toml

3. "配置 XXX 用某提供商的 API"

Read the provider's provider.toml for API endpoint:

gh api repos/anomalyco/models.dev/contents/providers/<provider-id>/provider.toml \
  --jq '.download_url' | xargs curl -sL

Key fields:

• api — OpenAI-compatible base URL
• npm — AI SDK package name (@ai-sdk/openai-compatible for generic OpenAI-compatible)
• env[] — environment variable names for auth

Then read the model TOML for model ID and capabilities. Combine into a configuration snippet for the user's target tool (Trae, OpenCode, Cursor, Continue, etc.).

4. "按条件筛选模型"

Use api.json with jq filtering:

# Find models under $X/M input that support reasoning + tool_call
curl -sL https://models.dev/api.json | jq '
  .[].models[] | select(.cost.input < 1 and .reasoning == true and .tool_call == true)
  | {id: .id, name, cost: .cost.input, context: .limit.context, reasoning}
'

Output Format

Present results clearly:

• Provider query: name, API endpoint, env vars, doc link
• Model list: table with Model ID, Name, Context, Input/M, Output/M, Capabilities
• Model detail: sectioned — Capabilities (reasoning/tool_call/structured_output/attachment/modalities), Limits (context/output), Pricing (input/output/cache_read), Interleaved (if reasoning model)

For Chinese-language queries, respond in Chinese. For English queries, respond in English — match the user's language.