MCPGen - Workflow Extension: leaderboard
Workflow extension in MCPGen (Yang et al., Shanghai Jiao Tong University, 2026): given the first half of a validated workflow as an immutable prefix, the tool schemas and a task description, the model writes the remaining steps and an executable runner for a Model Context Protocol project. Percent of the 100 tasks whose workflow and runner execute end to end in the project environment, single-turn at temperature 0; higher is better.
Metric: End-to-end success (%; share of the 100 extension tasks whose generated workflow and runner execute successfully in the project environment, given the first half of the gold workflow as an immutable prefix; single-turn generation, temperature 0, 100 executable MCP projects from 16 domains). Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 57 | #111 |
| 2 | O3 | 46 | #224 |
| 3 | Qwen 3 Next 80B A3B Instruct | 44 | #763 |
| 4 | Gemini 3 Flash (Preview) | 36 | #141 |
| 5 | MiniMax-M2.1 | 17 | #569 |
| 6 | Claude Sonnet 4.5 | 15 | #244 |
| 7 | Qwen 3 Coder 30B A3B Instruct | 15 | #973 |
| 8 | GPT-5.2 | 13 | #172 |
| 9 | GPT-5.1 | 13 | #225 |
| 10 | Claude Haiku 4.5 (20251001) | 10 | #529 |
| 11 | GPT-4o Mini | 6 | #1627 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mcpgen-workflow-extension · How It Works · Data refreshed daily, snapshot 2026-11-12.