API

以编程方式访问 AI 模型、基准测试和价格数据、开放 CSV 导出以及 Gecko Tests 结果。提供免费套餐。

对所有人免费 · 可用于个人、研究和商业用途

$0 · 无需 API 密钥

所有公开的 JSON 端点(模型、各提供商价格、基准测试、Gecko Tests、发现)
GitHub 上的 CSV 导出和每日数据集(CC BY 4.0)
面向 AI 助手和智能体的 MCP 服务器
价格历史和带日期的公司事实

唯一条件:在任何展示数据的地方标注“Source: BenchGecko”并链接到 benchgecko.ai(CC BY 4.0)。

免费层级归属要求

免费 API 访问要求在展示我们数据的页面上显示"Source: BenchGecko"链接。

<a href="https://benchgecko.ai">Source: BenchGecko</a>

端点

示例响应使用截至 2026年10月25日的真实 BenchGecko 数据,数组已缩短。

GET/api/v1/models

列出所有模型的分数和定价

参数

providerstring按提供商 slug 筛选(例如 "anthropic")

open_sourceboolean仅筛选开源模型

sortstring排序依据:avg_score、pricing_input、release_date、context_window

limitnumber每页结果数(默认 50,最大 200)

pagenumber页码(默认 1)

响应

{
  "data": [
    {
      "slug": "claude-mythos-preview",
      "name": "Claude Mythos Preview",
      "provider": "Anthropic",
      "provider_slug": "anthropic",
      "avg_score": 98.9,
      "scores": {
        "swe-bench-verified": 93.9,
        "swe-bench-pro": 77.8,
        "swe-bench-multilingual": 87.3
      },
      "pricing": {
        "input": null,
        "output": null
      },
      "context_window": 1000000,
      "is_open_source": false,
      "release_date": "2026-04-07",
      "model_type": "text"
    }
  ],
  "meta": {
    "total": 1271,
    "page": 1,
    "limit": 50,
    "pages": 26
  }
}

GET/api/v1/models/:slug

获取单个模型的详细数据

响应

{
  "slug": "claude-mythos-preview",
  "name": "Claude Mythos Preview",
  "provider": {
    "name": "Anthropic",
    "slug": "anthropic"
  },
  "avg_score": 98.9,
  "scores": [
    {
      "benchmark": "SWE-Bench verified",
      "benchmark_slug": "swe-bench-verified",
      "score": 93.9,
      "category": "coding"
    },
    {
      "benchmark": "SWE-bench Pro",
      "benchmark_slug": "swe-bench-pro",
      "score": 77.8,
      "category": "coding"
    }
  ],
  "pricing": {
    "input": null,
    "output": null
  },
  "context_window": 1000000,
  "is_open_source": false,
  "model_type": "text",
  "release_date": "2026-04-07"
}

GET/api/v1/benchmarks

列出所有基准测试及最高分模型

参数

categorystring按类别筛选:coding、reasoning、math、knowledge

响应

{
  "data": [
    {
      "slug": "gpqa-diamond",
      "name": "GPQA diamond",
      "category": "knowledge",
      "max_score": 100,
      "unit": "%",
      "models_tested": 178,
      "top_models": [
        {
          "name": "Gemini 4 Argon",
          "slug": "gemini-4-argon",
          "provider": "Google DeepMind",
          "score": 99.17
        },
        {
          "name": "GPT-6 Astra",
          "slug": "gpt-6-astra",
          "provider": "OpenAI",
          "score": 98.29
        }
      ]
    }
  ]
}

GET/api/v1/compare

两个或多个模型的正面对比

参数

modelsstring以逗号分隔的模型 slug,2 到 6 个(例如“claude-mythos-preview,gpt-5-5-pro”)

响应

{
  "models": [
    {
      "slug": "claude-mythos-preview",
      "name": "Claude Mythos Preview",
      "provider": "Anthropic",
      "avg_score": 98.9,
      "pricing": {
        "input": null,
        "output": null
      },
      "context_window": 1000000,
      "is_open_source": false
    },
    {
      "slug": "gpt-5-5-pro",
      "name": "GPT-5.5 Pro",
      "provider": "OpenAI",
      "avg_score": 98.3,
      "pricing": {
        "input": 30,
        "output": 180
      },
      "context_window": 400000,
      "is_open_source": false
    }
  ],
  "benchmarks": [
    {
      "name": "GPQA diamond",
      "slug": "gpqa-diamond",
      "category": "knowledge",
      "scores": {
        "claude-mythos-preview": 94.5,
        "gpt-5-5-pro": 94.2
      }
    }
  ]
}

GET/api/v1/export/:dataset

CSV 格式的开放数据(CC BY 4.0):price-history、provider-offers、price-changes、company-facts、mindshare-daily、gpu-rental-daily、world-map。仅包含 BenchGecko 自行收集的数据,不含第三方基准测试分数。

GET/api/lab/:test

JSON 格式的 Gecko Tests 结果:测试题目、每个模型的回答和分数。测试包括:who-are-you、tokenizer-tax、knowledge-horizon、censorship-index、model-drift、same-model-different-host。CC BY 4.0。

SDK、数据集与开源

官方客户端库、GitHub 上的每日开放数据集(CC BY 4.0)、Hugging Face、Kaggle 和 Harvard Dataverse 上的快照镜像,以及文档镜像。

探索开发者中心 →

速率限制

无需密钥,也没有硬性配额。响应会在边缘缓存最长一小时,请你也在本地做缓存。损害服务的流量可能会被屏蔽。

AI assistants and agents

BenchGecko is built to be quoted. Every JSON response includes source, url, as_of, license, attribution and a ready-made cite string. Model, test and scorecard pages link their JSON with <link rel="alternate" type="application/json">.

Remote MCP server

https://benchgecko.ai/api/mcp

Streamable HTTP, stateless, read-only, no key needed. Every tool result ends with “Source: BenchGecko · <url>”.

search_models · Search AI models
get_model · Get model facts
cheapest_provider · Cheapest provider for a model
compare_models · Compare two models
get_gecko_test · Gecko Test results
get_scorecard · Gecko Scorecard
latest_findings · Latest findings
search · Search BenchGecko
fetch · Fetch a BenchGecko record

Claude (claude.ai, Claude Desktop)

Settings · Connectors · Add custom connector, then paste the URL above. For older Claude Desktop builds, use the mcp-remote bridge in claude_desktop_config.json:

{
  "mcpServers": {
    "benchgecko": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://benchgecko.ai/api/mcp"
      ]
    }
  }
}

Claude Code

claude mcp add --transport http benchgecko https://benchgecko.ai/api/mcp

Cursor (.cursor/mcp.json) and other clients with remote MCP

{
  "mcpServers": {
    "benchgecko": {
      "url": "https://benchgecko.ai/api/mcp"
    }
  }
}

ChatGPT

Settings · Connectors · Create (developer mode), server URL above, no authentication. The search and fetch tools follow the ChatGPT connector format.

Test it

curl -s https://benchgecko.ai/api/mcp -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"cheapest_provider","arguments":{"model":"deepseek-v4-pro"}}}'

/openapi.json

OpenAPI 3.1 for every public read endpoint

/llms.txt

What BenchGecko is, how to cite it, live Gecko Tests snapshot

/llms-full.txt

Every test result, price spreads, companies, GPU rental, glossary

JSON with as-of dates and citations

/api/v1/models/<slug>One model: BenchGecko score and rank, list price, price at every provider, benchmark scores with their source, Gecko Tests grades, as-of dates

/api/v1/models?q=<name>Search the live catalog (or list with provider, open_source, sort)

/api/v1/compare?models=a,b2 to 6 models side by side

/api/v1/price-history/<slug>List price history (add ?envelope=1 for source and as-of fields)

/api/v1/gecko-testsEvery Gecko Test with its latest headline, the scorecard top line and the latest findings

/api/v1/gecko-tests/<test>Per-model results of one test, or the scorecard

/api/lab/<test>Full open dataset: items, runs and every model answer

/api/v1/findingsLatest notable results

/api/v1/export/<dataset>CSV exports (CC BY 4.0), mirrored daily to github.com/BenchGecko/datasets

License: data BenchGecko collects or measures itself (prices per provider, price history, Gecko Tests, mindshare, GPU rental) is CC BY 4.0. Benchmark scores are aggregated from public leaderboards and keep their original source. Credit “Source: BenchGecko” with a link.