ARC (AI2 Challenge) leaderboard
As of 6 Nov 2026, the top score on ARC (AI2 Challenge) is 95.3%, shared by DeepSeek V3 and Llama 3.1 405B, ahead of Qwen2.5 72B at 94.5%, across 77 model configurations with a published result (Epoch AI Benchmarking Hub, result dated 26 Dec 2024). Grade-school science questions, challenge set. ARC (AI2 Challenge) is shown for reference only and does not count toward the BenchLeader Index.
Key facts (as of 6 Nov 2026)
- DeepSeek V3 and Llama 3.1 405B share the top ARC (AI2 Challenge) score at 95.3% (Epoch AI Benchmarking Hub, 6 Nov 2026).
- ARC (AI2 Challenge): Grade-school science questions, challenge set. A 2018 benchmark, saturated for current models; kept for history (Epoch AI Benchmarking Hub).
- 77 model configurations have a published ARC (AI2 Challenge) result on BenchLeader (BenchLeader, 6 Nov 2026).
- ARC (AI2 Challenge) does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).
Grade-school science questions, challenge set. A 2018 benchmark, saturated for current models; kept for history.
- Published by
- Epoch AI Benchmarking Hub
- Category
- Reasoning
- Index weight
- Reference only
- Models
- 77
- Data as of
- 6 Nov 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
Grade-school science questions, challenge set.
How it is scored
Accuracy, as compiled by Epoch AI from published results.
What to keep in mind
Saturated: frontier models score near the ceiling, so it separates only older or smaller models, and results are largely self-reported by developers.
- 1DeepSeek V395.3%
- 2Llama 3.1 405B95.3%
- 3Qwen2.5 72B94.5%
- 4Deepseek v292.2%
- 5Phi 3 (medium)91.6%
- 6Phi 3 Small90.7%
- 7GPT 3.5 Turbo 110687.4%
- 8Mixtral 8x7B87.3%
- 9Claude Instant 1.286.3%
- 10Stable Beluga 286.1%
- 11Claude Instant 1.185.7%
- 12PaLM 540B85.2%
- 13text-davinci-00285.2%
- 14Phi 3 Mini84.9%
- 15Qwen-14B84.4%
| # | ||||
|---|---|---|---|---|
| 1 |  DeepSeek V3open ↗ | 95.3% | 44.4 | 2024-12-26 |
| 2 |  Llama 3.1 405Bopen ↗ | 95.3% | 41.9 | 2024-07-23 |
| 3 |  Qwen2.5 72Bopen ↗ | 94.5% | 43.2 | 2024-09-19 |
| 4 | Deepseek v2open ↗ | 92.2% | – | 2024-05-07 |
| 5 |  Phi 3medium | 91.6% | 37.9 | 2024-04-23 |
| 6 | Phi 3 Small | 90.7% | – | 2024-04-23 |
| 7 |  GPT 3.5 Turbo 1106 | 87.4% | 38.6 | 2023-11-06 |
| 8 |  Mixtral 8x7Bopen | 87.3% | 30.9 | 2023-12-11 |
| 9 |  Claude Instant 1.2 | 86.3% | – | 2023-08-09 |
| 10 |  Stable Beluga 2open | 86.1% | – | 2023-07-20 |
| 11 | Claude Instant 1.1 | 85.7% | – | – |
| 12 |  PaLM 540B | 85.2% | – | 2022-04-04 |
| 13 | text-davinci-002 | 85.2% | – | 2022-03-15 |
| 14 | Phi 3 Miniopen | 84.9% | 36.9 | 2024-04-23 |
| 15 | Qwen-14B | 84.4% | – | 2023-09-28 |
| 16 | Llama 3-8Bopen | 82.8% | 34.9 | 2024-04-18 |
| 17 |  InternLM 20B | 81.7% | – | 2023-09-18 |
| 18 | Mistral 7B v0.3open | 78.6% | 29.0 | 2023-09-27 |
| 19 | Llama 2-70Bopen ↗ | 78.3% | 33.6 | 2023-07-18 |
| 20 | Gemma 7B | 78.3% | – | 2024-02-21 |
| 21 | Phi-2open ↗ | 75.9% | – | 2023-12-12 |
| 22 | Qwen-7B | 75.3% | – | 2023-09-28 |
| 23 | Qwen2.5 Coder 32B Instructopen ↗ | 70.5% | 42.7 | 2024-09-18 |
| 24 | InternLM 7B | 69.5% | – | 2023-07-05 |
| 25 | LLaMA-65Bopen | 69.5% | – | 2023-02-24 |
| 26 | PaLM 2-L | 69.2% | – | 2023-05-17 |
| 27 |  Falcon-180B | 67.8% | – | 2023-09-06 |
| 28 | LLaMA-33B | 67.5% | – | 2023-02-24 |
| 29 | Qwen2.5-Coder-14B | 66.0% | – | 2024-09-18 |
| 30 | PaLM 2-M | 64.9% | – | 2023-05-17 |
| 31 | DeepSeek-Coder-V2 236B | 64.3% | – | 2024-06-17 |
| 32 | Falcon-40B | 61.9% | – | 2023-03-15 |
| 33 |  ChatGLM2-6B | 61.0% | – | 2023-06-24 |
| 34 | Qwen2.5-Coder 7B Instructopen ↗ | 60.9% | – | 2024-09-18 |
| 35 | Llama 2-13Bopen ↗ | 60.3% | – | 2023-07-18 |
| 36 | PaLM 2-S | 59.6% | – | 2023-05-17 |
| 37 | DeepSeek-Coder-V2-Lite-Base | 57.3% | – | 2024-06-13 |
| 38 |  Yi-9B | 55.6% | – | 2024-03-01 |
| 39 |  Nemotron-4 15B | 55.5% | – | 2024-02-26 |
| 40 |  INTELLECT-1 | 54.5% | – | 2024-11-29 |
| 41 | Llama 2-34B | 54.5% | – | 2023-07-18 |
| 42 | InstructGPT 175B | 53.2% | – | 2022-01-27 |
| 43 | Qwen-1_8B | 53.2% | – | 2023-11-30 |
| 44 | Qwen2.5-Coder-3B | 52.9% | – | 2024-09-18 |
| 45 | LLaMA-13B | 52.7% | – | 2023-02-24 |
| 46 | PaLM 62B | 52.5% | – | 2022-04-04 |
| 47 |  MPT-30B | 50.6% | – | 2023-06-22 |
| 48 | Yi 6B | 50.3% | – | 2023-11-02 |
| 49 | Falcon-7B | 47.9% | – | 2023-04-24 |
| 50 | LLaMA-7B | 47.6% | – | 2023-02-24 |
| 51 |  StarCoder2-15Bopen | 47.2% | – | 2024-02-20 |
| 52 | Llama 2-7Bopen ↗ | 45.9% | – | 2023-07-18 |
| 53 | Qwen2.5 Coder 1.5B | 45.2% | – | 2024-09-18 |
| 54 | Phi-1.5 | 44.4% | – | 2023-09-11 |
| 55 |  Vicuna 13B v1.1 | 43.2% | – | 2023-04-12 |
| 56 | MPT-7B | 42.6% | – | 2023-05-05 |
| 57 | Deepseek Coder 33B Base | 42.2% | – | 2023-11-02 |
| 58 | Gemma 2B | 42.1% | – | 2024-02-21 |
| 59 |  Xgen 7B Base | 41.2% | – | 2023-06-27 |
| 60 |  GPT-NeoX-20B | 41.1% | – | 2022-04-07 |
More reasoning benchmarks · All LLM benchmarks
Related benchmarks
Cite as: BenchLeader, “ARC (AI2 Challenge) leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/arc_ai2.
ARC (AI2 Challenge): questions
- What does ARC (AI2 Challenge) measure?
- Grade-school science questions, challenge set. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on ARC (AI2 Challenge)?
- As of 6 Nov 2026, the highest score on ARC (AI2 Challenge) is 95.3%, shared by DeepSeek V3 and Llama 3.1 405B, ahead of Qwen2.5 72B at 94.5%. Scores come from Epoch AI Benchmarking Hub and are re-read daily.
- Which AI model leads ARC (AI2 Challenge)?
- DeepSeek V3 and Llama 3.1 405B tie for first on ARC (AI2 Challenge) at 95.3% as of 6 Nov 2026, ahead of Qwen2.5 72B at 94.5%.
- How many models have ARC (AI2 Challenge) results?
- 77 model configurations have a ARC (AI2 Challenge) result on BenchLeader, all taken from Epoch AI Benchmarking Hub.
- Who runs ARC (AI2 Challenge) and how often is it updated?
- ARC (AI2 Challenge) is published by Epoch AI Benchmarking Hub. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does ARC (AI2 Challenge) count toward the BenchLeader Index?
- No. ARC (AI2 Challenge) is shown for reference but left out of the composite index.