ARC-AGI-2
Abstract visual puzzles that are easy for people and hard for models.
Agent Index weight
0%
shown for context only
Models tested
96
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | GPT-6 Astra Supported OpenAI | 95.0% | source | — | — | |
| 2 | Gemini 4 Carbon Supported Google | 94.2% | source | — | — | |
| 3 | GPT-6.1 Sol Estimated OpenAI | 94.2% | source | — | — | |
| 4 | Claude Fable 5.2 Supported Anthropic | 93.8% | source | — | — | |
| 5 | Claude Opus 5.5 Supported Anthropic | 93.3% | source | — | — | |
| 6 | GPT-5.6 Sol Supported OpenAI | 92.5% | source | $1.44 | — | |
| 7 | Gemini 4 Argon Supported Google | 91.3% | source | — | — | |
| 8 | Claude Opus 5 Supported Anthropic | 90.4% | source | — | — | |
| 9 | Claude Fable 5.1 Supported Anthropic | 90.0% | source | — | — | |
| 10 | GPT-6 Sol Estimated OpenAI | 89.6% | source | — | — | |
| 11 | Claude Fable 5 Supported Anthropic | 89.2% | source | — | — | |
| 12 | Gemini 3.8 Flash Supported Google | 89.2% | source | — | — | |
| 13 | GPT-5.5 Supported OpenAI | 85.0% | source | $1.87 | — | |
| 14 | Gemini 3.7 Flash Estimated Google | 84.6% | source | — | — | |
| 15 | Gemini 3 Deep Think Google | 84.6% | source | — | — | |
| 16 | GPT-5.5 Pro OpenAI | 84.6% | source | $10.51 | — | |
| 17 | GPT-5.6 Terra Estimated OpenAI | 83.9% | source | $1.09 | — | |
| 18 | Claude Sonnet 5.5 Estimated Anthropic | 83.9% | source | — | — | |
| 19 | GPT-5.4 Pro OpenAI | 83.3% | source | $16.41 | — | |
| 20 | Inkling 2 Small Estimated thinkingmachines | 81.7% | source | — | — | |
| 21 | DeepSeek V4.1 Flash 1017 Estimated DeepSeek | 78.1% | source | — | — | |
| 22 | Gemini 3.1 Pro Preview Supported Google | 77.1% | source | $0.96 | — | |
| 23 | Claude Opus 4.7 Supported Anthropic | 75.8% | source | — | — | |
| 24 | GPT-5.4 Supported OpenAI | 74.0% | source | $1.52 | — | |
| 25 | DeepSeek V4.1 Flash Estimated DeepSeek | 72.9% | source | — | — | |
| 26 | Claude Opus 4.8 Supported Anthropic | 72.1% | source | $2.74 | — | |
| 27 | Gemini 3.5 Flash Supported Google | 72.1% | source | $0.85 | — | |
| 28 | Claude Opus 4.6 Supported Anthropic | 69.2% | source | — | — | |
| 29 | Grok 4.6 Supported xAI | 67.1% | source | — | — | |
| 30 | GLM-5.3-Flash Supported Z.ai (Zhipu) | 65.8% | source | — | — | |
| 31 | Grok 4.20 (Non-Reasoning) Estimated xAI | 65.1% | source | $0.92 | — | |
| 32 | DeepSeek V4 Flash 0731 DeepSeek | 61.4% | source | — | — | |
| 33 | Grok 4.7 Estimated xAI | 61.4% | source | — | — | |
| 34 | Claude Haiku 5.5 Estimated Anthropic | 61.3% | source | — | — | |
| 35 | DeepSeek V4 Pro 0813 DeepSeek | 61.3% | source | — | — | |
| 36 | Claude Sonnet 4.6 Supported Anthropic | 60.4% | source | $2.70 | — | |
| 37 | Gemini 3.6 Flash Estimated Google | 60.4% | source | — | — | |
| 38 | Kimi K3 Supported Moonshot AI | 60.4% | source | $1.59 | — | |
| 39 | GPT-5.6 Luna Estimated OpenAI | 59.5% | source | $0.67 | — | |
| 40 | GPT-6 Luna OpenAI | 59.3% | source | — | — | |
| 41 | Muse Spark Open Meta | 57.5% | source | — | — | |
| 42 | GPT-5.2 Pro OpenAI | 54.2% | source | $15.72 | — | |
| 43 | GPT-5.2 Supported OpenAI | 52.9% | source | $1.90 | — | |
| 44 | Grok 4.5 Estimated xAI | 52.6% | source | $0.78 | — | |
| 45 | Qwen3.8 27B Alibaba (Qwen) | 42.4% | source | — | — | |
| 46 | Inkling Small thinkingmachines | 40.1% | source | — | — | |
| 47 | Claude Opus 4.5 (latest) Supported Anthropic | 37.6% | source | $2.40 | — | |
| 48 | Inkling Estimated thinkingmachines | 36.5% | source | $0.64 | — | |
| 49 | Gemini 3 Flash Preview Supported Google | 33.6% | source | — | — | |
| 50 | Gemini 3 Pro Preview Supported Google | 31.1% | source | — | — | |
| 51 | Beam 1.0 Reflection | 23.5% | source | — | — | |
| 52 | GLM-5.2 Supported Z.ai (Zhipu) | 22.8% | source | $0.25 | — | |
| 53 | GPT-5.4 mini OpenAI | 18.9% | source | $0.75 | — | |
| 54 | GPT-5 Pro OpenAI | 18.3% | source | $7.14 | — | |
| 55 | GPT-5.1 Supported OpenAI | 17.6% | source | $1.17 | — | |
| 56 | Grok 4 Supported xAI | 16.0% | source | — | — | |
| 57 | Claude Sonnet 4.5 (latest) Supported Anthropic | 13.6% | source | $0.76 | — | |
| 58 | Kimi K2.5 Supported Moonshot AI | 11.8% | source | $0.28 | — | |
| 59 | Gemini 3.5 Flash Lite Google | 10.3% | source | — | — | |
| 60 | GPT-5 Supported OpenAI | 9.9% | source | $0.73 | — | |
| 61 | Claude-Opus-4 Supported Anthropic | 8.6% | source | $1.93 | — | |
| 62 | o3 Supported OpenAI | 6.5% | source | $0.83 | — | |
| 63 | tiny-recursion-model Unknown | 6.3% | source | $2.10 | — | |
| 64 | o4-mini Estimated OpenAI | 6.1% | source | $0.86 | — | |
| 65 | Claude Sonnet 4 Supported Anthropic | 5.9% | source | $0.49 | — | |
| 66 | GPT-5.4 nano OpenAI | 5.7% | source | $0.16 | — | |
| 67 | Grok 4 Fast xAI | 5.3% | source | $0.06 | — | |
| 68 | Gemini 2.5 Pro Supported Google | 4.9% | source | $0.76 | — | |
| 69 | GLM-5 Supported Z.ai (Zhipu) | 4.9% | source | $0.27 | — | |
| 70 | MiniMax-M2.5 Estimated MiniMax | 4.9% | source | $0.17 | — | |
| 71 | o3-pro OpenAI | 4.9% | source | $7.55 | — | |
| 72 | GPT-5 Mini Estimated OpenAI | 4.4% | source | $0.20 | — | |
| 73 | Claude Haiku 4.5 Estimated Anthropic | 4.0% | source | $0.38 | — | |
| 74 | DeepSeek V3.2 Supported DeepSeek | 4.0% | source | $0.12 | — | |
| 75 | o3-mini Estimated OpenAI | 3.0% | source | $0.55 | — | |
| 76 | GPT-5 Nano OpenAI | 2.6% | source | $0.03 | — | |
| 77 | Gemini 2.5 Flash Estimated Google | 2.5% | source | $0.32 | — | |
| 78 | DeepSeek-R1 Estimated DeepSeek | 1.3% | source | $0.08 | — | |
| 79 | Gemini-2.0-Flash Google | 1.3% | source | $0.00 | — | |
| 80 | codex-mini-2025-05-16 Unknown | 1.3% | source | $0.23 | — | |
| 81 | Qwen3 235B A22B Instruct 2507 Alibaba (Qwen) | 1.3% | source | $0.00 | — | |
| 82 | Claude 3.7 Sonnet Supported Anthropic | 0.9% | source | $0.36 | — | |
| 83 | o1-mini OpenAI | 0.8% | source | $0.19 | — | |
| 84 | Gemini 1.5 Pro (Feb 2024) Google | 0.8% | source | $0.04 | — | |
| 85 | GPT-4.5 OpenAI | 0.8% | source | $2.10 | — | |
| 86 | GPT-4.1 OpenAI | 0.4% | source | $0.07 | — | |
| 87 | Grok-3 mini xAI | 0.4% | source | $0.01 | — | |
| 88 | GPT-4.1 mini OpenAI | 0.0% | source | $0.01 | — | |
| 89 | GPT-4.1 nano OpenAI | 0.0% | source | $0.00 | — | |
| 90 | GPT-4o (2024-08-06) Supported OpenAI | 0.0% | source | $0.08 | — | |
| 91 | GPT-4o mini OpenAI | 0.0% | source | $0.01 | — | |
| 92 | Grok 3 xAI | 0.0% | source | $0.14 | — | |
| 93 | Llama 4 Maverick Meta | 0.0% | source | $0.01 | — | |
| 94 | Llama 4 Scout Meta | 0.0% | source | $0.01 | — | |
| 95 | Magistral Medium 1.1 Mistral | 0.0% | source | — | — | |
| 96 | Magistral Small 1.0 Mistral | 0.0% | source | $0.05 | — |