GAIA2: leaderboard
Meta's Agents Research Environments (ARE) benchmark scoring general assistants on execution, search, ambiguity, adaptability, and robustness to noise.
Metric: Pass@1 (%). Source: huggingface.co. Saturation forecast: Around January 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (High) | 42.1 | #98 (GPT-5) |
| 2 | Claude Sonnet 4 (Thinking) | 37.8 | #199 (Claude Sonnet 4) |
| 3 | Claude Sonnet 4 | 34.8 | #199 |
| 4 | GPT-5 (Low) | 34.6 | #98 (GPT-5) |
| 5 | Gemini 2.5 Pro | 25.8 | #151 |
| 6 | DeepSeek V3.1 Terminus | 23.1 | #219 |
| 7 | DeepSeek V3.1 | 21.9 | #264 |
| 8 | Kimi K2 | 20.1 | #243 |
| 9 | GPT-5 (Minimal) | 18.2 | #98 (GPT-5) |
| 10 | Grok 4 | 15.7 | #181 |
| 11 | Qwen 3 235B A22B (Thinking) | 15.7 | #312 (Qwen 3 235B A22B) |
| 12 | GPT-OSS-120B (High) | 13.7 | #341 (GPT-OSS-120B) |
| 13 | Qwen 3 235B A22B | 11.6 | #312 |
| 14 | GPT-4o | 7.4 | #343 |
| 15 | Llama 4 Maverick | 7.4 | #466 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=gaia2 · How It Works · Data refreshed daily, snapshot 2026-10-21.