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. Status: saturation imminent. 39 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 (High) | 57.48 |
| 2 | Gemini 3 Pro (High) | 53.94 |
| 3 | GPT-5 (Medium) | 52.56 |
| 4 | GPT-5 (High) | 51.57 |
| 5 | GPT-5 (Low) | 46.85 |
| 6 | Claude Opus 4.5 (High) | 42.32 |
| 7 | DeepSeek V3.2 (Thinking) | 36.81 |
| 8 | Claude Sonnet 4.5 | 32.09 |
| 9 | Grok 4 | 31.69 |
| 10 | GPT-5 Mini (High) | 30.32 |
| 11 | Claude Opus 4.1 | 29.92 |
| 12 | Claude Sonnet 4 | 28.15 |
| 13 | GPT-5 Mini (Medium) | 27.36 |
| 14 | O3 | 25.39 |
| 15 | Qwen 3 Coder Plus | 24.8 |
Interactive version: theaggregate.ai/benchmark?slug=mcpmark · How the rankings work · Data refreshed daily, snapshot 2026-07-22.