MCP-Persona: leaderboard
MCP-Persona (ICML 2026) gives tool-calling agents personalized tasks, such as messaging a colleague found through a shared group, on 12 MCP servers simulated from real tool traces, with up to 20 rounds per task. The board is checkpoint accuracy over all 173 tasks: each sub-task checkpoint is scored 0, 0.5 or 1 by a GPT-4o judge and averaged, in percent; higher is better.
Metric: Checkpoint accuracy (%): mean LLM-judged score (0, 0.5 or 1) of the sub-task checkpoints in a task, averaged over tasks, all tasks; 173 human-verified personal-application tasks on 12 simulated MCP servers (Lark, Slack, WeCom, Notion, Obsidian, Rednote, Reddit, Instagram, email and search tools), up to 20 tool-calling rounds, GPT-4o judge; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 38.66 | #250 |
| 2 | GPT-5 | 36.99 | #167 |
| 3 | Claude Opus 4.1 | 34.52 | #272 |
| 4 | O4 Mini | 30.7 | #353 |
| 5 | O3 | 29.79 | #228 |
| 6 | Qwen 3 235B A22B 2507 Instruct | 26.75 | #619 |
| 7 | GPT-4o (2024-11-20) | 25.56 | #866 |
| 8 | DeepSeek V3 (0324) | 25.29 | #709 |
| 9 | Grok 4 | 24.58 | #307 |
| 10 | Qwen3 Coder | 23.93 | #586 |
| 11 | Gemini 2.5 Pro | 20.68 | #266 |
| 12 | Gemini 3 Pro | 16.91 | #137 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mcp-persona · How It Works · Data refreshed daily, snapshot 2026-11-13.