OlympicArena: leaderboard
OlympicArena — Olympiad-level multidisciplinary benchmark (math, physics, chemistry, biology, CS and more) drawn from international olympiad problems, testing complex and multimodal reasoning.
Metric: Overall Accuracy (%). Source: gair-nlp.github.io. Saturation forecast: Estimated already saturated. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 34.01 | #343 |
| 2 | Claude 3.5 Sonnet | 33.54 | #350 |
| 3 | Qwen 2 72B Instruct | 28.84 | #576 |
| 4 | Gemini 1.5 Pro | 28.68 | #445 |
| 5 | GPT-4o Mini | 28.37 | #599 |
| 6 | Llama 3 70B Instruct | 21.63 | #634 |
| 7 | Qwen 1.5 32B Chat | 21.16 | #816 |
| 8 | Gemma 2 27B (IT) | 19.91 | #645 |
| 9 | Claude 3 Sonnet | 18.97 | #775 |
| 10 | InternLM2.5-Chat-7B | 18.65 | #1034 |
| 11 | Llama 3.1 8B Instruct | 14.73 | #1035 |
| 12 | InternVL2-8B | 14.73 | #817 |
| 13 | internlm2-chat-20B | 13.95 | #999 |
| 14 | Yi Chat 34B | 12.54 | #976 |
| 15 | Qwen-7B Chat | 3.45 | #1380 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=olympicarena · How It Works · Data refreshed daily, snapshot 2026-10-20.