CAIA - Pass@1 (With Tools): leaderboard
The with-tools track of CAIA (Crypto AI Agent benchmark) — agents tackle high-stakes crypto and web3 analyst tasks in an adversarial setting with access to tools; Pass@1 (%).
Metric: Pass@1 (%). Source: huggingface.co. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 70.2 | #98 |
| 2 | O3 | 59.6 | #124 |
| 3 | Claude Opus 4 | 59.6 | #163 |
| 4 | Claude Sonnet 4 | 57.9 | #199 |
| 5 | Grok 4 Fast (Reasoning) | 57.9 | #254 (Grok 4 Fast) |
| 6 | Claude Opus 4.1 | 56.3 | #140 |
| 7 | GPT-OSS-120B | 56.2 | #341 |
| 8 | DeepSeek V3.1 | 55.9 | #264 |
| 9 | Grok 4 | 52.2 | #181 |
| 10 | GPT-4.1 | 51.7 | #248 |
| 11 | GPT-4o | 50 | #343 |
| 12 | Gemini 2.5 Pro | 49.4 | #152 |
| 13 | Kimi K2 | 46.6 | #243 |
| 14 | Gemini 2.5 Flash | 39.3 | #245 |
| 15 | Qwen 3 235B A22B | 38.4 | #313 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=caia-pass-1-with-tools · How It Works · Data refreshed daily, snapshot 2026-10-20.