MCPMark-Verified
Agents
MCPMark-Verified is a human-verified edition of MCPMark, which evaluates agents using Model Context Protocol tools against five real server environments (Notion, GitHub, Filesystem, PostgreSQL and Playwright) with programmatic verification of each task's end state. Releases report pass@1 averaged over runs under the official configuration (100-turn tool-call budget); the verified task roster had not been published when it was first reported, so no item count is declared. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K3 | 94.5 |
| GPT-5.5 | 92.9 |
| GPT-5.6 Sol | 92.9 |
| Kimi K2.7 Code | 81.1 |
| Opus 4.8 | 76.4 |
| Kimi K2.6 | 72.8 |