All Benchmarks/Agents

MCPMark-Verified

Often citedHigher is better

The verified edition of MCPMark, which tests whether the AI can complete tasks using real MCP servers such as GitHub, Notion, and PostgreSQL. The maintainers re-checked the tasks, so scores are not comparable with the earlier edition. Scores run from 0 to 100%. Higher is better.

Other versions:MCPMark
Top score96.1%Kimi K3
Models tested8
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    Moonshot AI
    Kimi K3
    2026.07.17 · max
    96.1%
  2. 2
    OpenAI
    GPT-5.5
    2026.04.23 · xhigh
    92.9%
  3. 2
    OpenAI
    GPT-5.6 Sol
    2026.07.09 · max
    92.9%
  4. 4
    Anthropic
    Claude Fable 5
    2026.06.09 · max
    86.6%
  5. 5
    Moonshot AI
    Kimi K2.7 Code
    2026.06.12
    81.9%
  6. 6
    Anthropic
    Claude Opus 4.8
    2026.05.27 · max
    76.4%
  7. 7
    Moonshot AI
    Kimi K2.6
    2026.04.20
    72.8%
  8. 8
    DeepSeek
    DeepSeek V4 Pro
    2026.04.24 · max
    71.7%

The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.