EnterpriseBench Corecraft (Paper Task Set): leaderboard
Metric: Task pass rate (%) on the Corecraft benchmark task set evaluated in the paper (expert-written customer-support tasks in a simulated PC-parts retailer with 23 MCP tools); a task passes only when an LLM judge finds every expert-authored rubric criterion satisfied across its trajectories; higher is better. Source: arxiv.org. Saturation forecast: Around October 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort) | 30.8 | #60 (Claude Opus 4.6) |
| 2 | GPT-5.2 (High) | 29.7 | #105 (GPT-5.2) |
| 3 | Gemini 3.1 Pro (Preview) | 27.2 | #54 |
| 4 | Claude Opus 4.6 (High) | 26.2 | #60 (Claude Opus 4.6) |
| 5 | DeepSeek V3.2 (Thinking) | 24.1 | #198 (DeepSeek V3.2) |
| 6 | Claude Opus 4.6 | 22.1 | #60 |
| 7 | Grok 4.1 Fast (Reasoning) | 20.5 | #208 (Grok 4.1 Fast) |
| 8 | GPT-5.2 Codex (xHigh) | 20.1 | #89 (GPT-5.2 Codex) |
| 9 | Gemini 3 Flash | 20 | #93 |
| 10 | GLM-5 | 17.4 | #137 |
| 11 | Claude Sonnet 4.6 | 16.4 | #85 |
| 12 | Gemini 3 Pro | 14.4 | #77 |
| 13 | Qwen 3.5 Plus | 11.3 | #123 |
| 14 | Kimi K2.5 | 8.7 | #139 |
| 15 | Mistral Large 3 | 3.6 | #388 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=enterprisebench-corecraft-paper-task-set · How It Works · Data refreshed daily, snapshot 2026-10-11.