ACE-Bench (Azure SDK, MCP Documentation): leaderboard
Metric: Strict pass rate (%) on ACE-Bench (353 Azure SDK coding tasks in Java, JavaScript/TypeScript, C# and Python built from official documentation examples; a task passes only if every atomic criterion holds: regex checks of required API usage and reference-grounded LLM-judge checks); same lightweight mcp-use coding agent for every model, augmented: the agent may query the Microsoft Learn MCP documentation server before answering; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Grok 4 | 68.7 |
| 2 | Claude Opus 4.5 | 65.3 |
| 3 | Claude Sonnet 4.5 | 63.7 |
| 4 | GPT-5.1 | 63.2 |
| 5 | Grok Code Fast 1 | 58.5 |
| 6 | Claude Haiku 4.5 | 58 |
| 7 | Claude Opus 4.1 | 53.6 |
| 8 | GPT-5 | 53.5 |
| 9 | GPT-4.1 | 51.1 |
| 10 | Grok 4 Fast (Non-reasoning) | 50.9 |
| 11 | GPT-5 Mini | 49.6 |
Interactive version: theaggregate.ai/benchmark?slug=ace-bench-azure-sdk-mcp-documentation · How It Works · Data refreshed daily, snapshot 2026-10-07.