ContractBench: leaderboard
Tool-augmented LLM agents call APIs whose intermediate outputs, such as presigned URLs, session tokens, and OAuth state parameters, are observation contracts: artifacts whose later use is constrained by the external system that produced them. We show that observation contract compliance (preserving the temporal validity and byte-level integrity) is an emergent, regression-prone capability: it is n
Metric: SR% (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 77.8 | #63 |
| 2 | GPT-5.2 | 74.8 | #110 |
| 3 | GPT-5 | 70.7 | #98 |
| 4 | Qwen 3.5 397B A17B | 70.7 | #146 |
| 5 | Claude Sonnet 4.5 | 69.7 | #142 |
| 6 | Qwen 3.5 27B | 64.6 | #187 |
| 7 | MiniMax-M2.5 | 62.6 | #303 |
| 8 | MiniMax-M2.1 | 60.6 | #289 |
| 9 | Qwen 3.5 9B | 56.6 | #379 |
| 10 | MiniMax-M2 | 53.5 | #317 |
| 11 | Gemini 2.5 Pro | 51.5 | #152 |
| 12 | GPT-5.1 | 48.5 | #135 |
| 13 | Mistral Small 4 | 42.4 | #517 |
| 14 | Gemma 4 26B A4B | 38.4 | #358 |
| 15 | Gemma 4 31B | 37.4 | #200 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=contractbench · How It Works · Data refreshed daily, snapshot 2026-10-20.