MCPEvol-Bench - Planning Effectiveness (Late Stage): leaderboard
MCPEvol-Bench (2026) tool-use tasks run on the servers after five evolution rounds: planning effectiveness rated 1 to 10 by a DeepSeek-Chat judge from the instruction, final answer and trajectory. Higher is better.
Metric: Planning Effectiveness score (1 to 10 rubric; servers after five evolution rounds; 201 synthesized multi-server tasks over a fixed set of candidate MCP servers from 123 real servers; rubric scores on a 1 to 10 scale from a DeepSeek-Chat judge over the instruction, final answer and execution trajectory (MCP-Bench evaluation framework)). Source: arxiv.org. Saturation forecast: Not forecast. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 31B (IT) | 6.68 | #342 |
| 2 | Claude Sonnet 4.6 | 6.17 | #161 |
| 3 | Claude Opus 4.6 | 6.03 | #102 |
| 4 | Qwen 3.5 27B | 5.12 | #359 |
| 5 | Claude Sonnet 4.5 (Thinking) | 4.98 | #250 (Claude Sonnet 4.5) |
| 6 | Qwen 3.5 9B | 4.86 | #803 |
| 7 | Llama 3.3 70B | 3.94 | #1498 |
| 8 | GPT-5.4 | 3.87 | #123 |
| 9 | Gemini 2.5 Pro | 3.71 | #266 |
| 10 | GPT-5.1 | 3.5 | #229 |
| 11 | O3 | 3.1 | #228 |
| 12 | GPT-4o | 2.78 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mcpevol-bench-planning-effectiveness-late-stage · How It Works · Data refreshed daily, snapshot 2026-11-13.