ARC Challenge (AI2): leaderboard
AI2 Reasoning Challenge: 7,787 science exam questions at grade-school level. The 'Challenge' set contains questions that simple retrieval and co-occurrence methods fail on.
Metric: Accuracy (%). Source: epoch.ai. Saturation forecast: Estimated already saturated. 77 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 | 95.3 | #321 |
| 2 | Llama 3.1 405B | 95.3 | #466 |
| 3 | Qwen 2.5 72B | 94.5 | #499 |
| 4 | DeepSeek V2 | 92.2 | #549 |
| 5 | Phi-3-medium-128k-instruct | 91.6 | #1013 |
| 6 | Phi-3-small-8k-instruct | 90.7 | #967 |
| 7 | GPT-3.5 Turbo (1106) | 87.4 | #961 |
| 8 | Mixtral 8x7B (v0.1) | 87.3 | #1083 |
| 9 | Claude Instant 1.2 | 86.3 | #1017 |
| 10 | StableBeluga2 | 86.1 | #1078 |
| 11 | text-davinci-002 | 85.2 | #1225 |
| 12 | Phi-3 Mini 4K Instruct | 84.9 | #1297 |
| 13 | Qwen-14B | 84.4 | #1171 |
| 14 | Llama 3 8B Instruct | 82.8 | #1129 |
| 15 | internlm-20B | 81.7 | #1296 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=arc-challenge-ai2 · How It Works · Data refreshed daily, snapshot 2026-10-20.