FactArena: leaderboard
Metric: Elo rating from pairwise judged comparisons. Source: arxiv.org. Saturation forecast: Around September 2026. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 (2025-04-16) | 1320.14 |
| 2 | DeepSeek R1 | 1127.77 |
| 3 | O4 Mini (2025-04-16) | 1108.47 |
| 4 | Gemini 2.5 Pro (Preview 06-05) | 1084.19 |
| 5 | GPT-4.5 (Preview) | 1047.15 |
| 6 | GPT-4.1 | 1041.71 |
| 7 | Grok 3 | 1022.09 |
| 8 | Grok 3 Mini Beta | 960.74 |
| 9 | Gemini 2.5 Flash | 955.6 |
| 10 | Claude Opus 4 (20250514) | 949.27 |
| 11 | DeepSeek V3 | 934.38 |
| 12 | Llama 4 Maverick Instruct | 919.57 |
| 13 | Claude Sonnet 4 (20250514) | 905.28 |
| 14 | Qwen 3 235B A22B | 890.89 |
| 15 | GPT-4o | 890.75 |
Interactive version: theaggregate.ai/benchmark?slug=factarena · How It Works · Data refreshed daily, snapshot 2026-09-25.