MCPGen - Tool Creation (Collective Pass): leaderboard
Tool creation in MCPGen (Yang et al., Shanghai Jiao Tong University, 2026), scored at project level: all generated tools of a Model Context Protocol project are substituted together and the project integration tests are run. Percent of projects that pass, single-turn at temperature 0; higher is better.
Metric: Collective pass rate (%; share of projects whose integration tests pass with every generated tool substituted at once; single-turn generation, temperature 0, 100 executable MCP projects from 16 domains). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 (20251001) | 45 | #529 |
| 2 | Claude Sonnet 4.5 | 41 | #244 |
| 3 | O3 | 39 | #224 |
| 4 | Qwen 3 Coder 30B A3B Instruct | 38 | #973 |
| 5 | GPT-5.1 | 37 | #225 |
| 6 | MiniMax-M2.1 | 37 | #569 |
| 7 | Qwen 3 Next 80B A3B Instruct | 36 | #763 |
| 8 | GPT-5.2 | 34 | #172 |
| 9 | Gemini 3 Pro (Preview) | 34 | #111 |
| 10 | GPT-4o Mini | 33 | #1627 |
| 11 | Gemini 3 Flash (Preview) | 32 | #141 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mcpgen-tool-creation-collective-pass · How It Works · Data refreshed daily, snapshot 2026-11-12.