MCPMark: leaderboard
MCP-based agent benchmark evaluating language-model agents across real-world tool environments including filesystem, GitHub, Notion, browser automation, and Postgres tasks.
Metric: Pass@1 (%). Source: mcpmark.ai. Saturation forecast: Around December 2026. 39 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 (High) | 57.48 | #110 (GPT-5.2) |
| 2 | Gemini 3 Pro (High) | 53.94 | #80 (Gemini 3 Pro) |
| 3 | GPT-5 (Medium) | 52.56 | #98 (GPT-5) |
| 4 | GPT-5 (High) | 51.57 | #98 (GPT-5) |
| 5 | Gemini 3 Pro (Low) | 50.79 | #80 (Gemini 3 Pro) |
| 6 | GPT-5 (Low) | 46.85 | #98 (GPT-5) |
| 7 | Claude Opus 4.5 (High) | 42.32 | #83 (Claude Opus 4.5) |
| 8 | DeepSeek V3.2 (Thinking) | 36.81 | #205 (DeepSeek V3.2) |
| 9 | Claude Sonnet 4.5 | 32.09 | #142 |
| 10 | Grok 4 | 31.69 | #181 |
| 11 | GPT-5 Mini (High) | 30.32 | #186 (GPT-5 Mini) |
| 12 | Claude Opus 4.1 | 29.92 | #140 |
| 13 | Claude Sonnet 4 | 28.15 | #199 |
| 14 | GPT-5 Mini (Medium) | 27.36 | #186 (GPT-5 Mini) |
| 15 | O3 | 25.39 | #124 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mcpmark · How It Works · Data refreshed daily, snapshot 2026-10-20.