Back to BenchmarksFrontier
ARC AI2
AI2 Reasoning Challenge - science questions requiring reasoning
Leaderboard
(20 models)| Rank | Model | Score | Stderr |
|---|---|---|---|
| 1 | Llama 3.1 405B | 95.30 | — |
| 2 | DeepSeek V3 | 95.30 | — |
| 3 | Qwen 2.5 72B | 94.50 | — |
| 4 | Phi-3-medium-128k-instruct | 91.60 | — |
| 5 | Phi-3-small-8k-instruct | 90.70 | — |
| 6 | gpt-3.5-turbo-1106 | 87.40 | — |
| 7 | Mixtral-8x7B-v0.1 | 87.30 | — |
| 8 | Claude Opus 4.5 | 86.30 | — |
| 9 | Phi-3-mini-4k-instruct | 84.90 | — |
| 10 | Qwen 3 235B | 84.40 | — |
| 11 | Meta-Llama-3-8B-Instruct | 82.80 | — |
| 12 | Mistral-7B-v0.1 | 78.60 | — |
| 13 | gemma-7b | 78.30 | — |
| 14 | Llama-2-70b-hf | 78.30 | — |
| 15 | Phi-4 | 75.90 | — |
| 16 | Qwen2.5-Max | 70.50 | — |
| 17 | falcon-180B | 67.80 | — |
| 18 | Llama-2-7b | 60.30 | — |
| 19 | Yi-6B | 55.60 | — |
| 20 | GPT-OSS 120B | 41.10 | — |
Data source: Epoch AI, “Data on AI Benchmarking”. Published at epoch.ai
Licensed under CC-BY 4.0