GAIA: leaderboard
General AI Assistants benchmark - 466 questions requiring multi-step reasoning, web browsing, and tool use. Questions are easy for humans (~92%) but hard for AI. Three difficulty levels.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around January 2027. 3802 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 77.41 | #110 |
| 2 | DeepSeek V3.2 | 76.74 | #205 |
| 3 | GPT-5 | 66.45 | #98 |
| 4 | Gemini 3.1 Pro (Preview) | 55.48 | #58 |
| 5 | Gemini 3 Flash | 41.53 | #99 |
| 6 | GPT-4o (2024-08-06) | 28.9 | #336 |
| 7 | DeepSeek V4 Flash | 23.59 | #137 |
| 8 | GPT-5 Nano | 23.26 | #425 |
| 9 | GPT-5.1 | 19.6 | #135 |
| 10 | GPT-4 Turbo | 11.3 | #404 |
| 11 | Gemma 4 26B A4B (IT) | 8.31 | #297 |
| 12 | GPT-4o Mini (2024-07-18) | 4.65 | #558 |
| 13 | GPT-4 | 3.99 | #438 |
| 14 | GPT-3.5 | 2.66 | #818 |
| 15 | Qwen 2.5 7B | 0.66 | #922 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=gaia · How It Works · Data refreshed daily, snapshot 2026-10-20.