ReasoningOfficial + Vendor scoresPrices verified
ARC-AGI-2 leaderboard
As of November 8, 2026, GPT-6 Astra (OpenAI) leads ARC-AGI-2 at 95%, ahead of GPT-6.1 Sol at 94.2% and Claude Opus 5.5 at 93.3%. 105 systems ranked. All 105 scores on the main board come from ARC Prize Foundation, not from the labs that make the models.
Top score
95%
Systems ranked
105
Independent rows
105/109
Cheapest near top
$1.21/1M
With API prices
72/105
Newest score
Nov 5
What is ARC-AGI-2? ARC-AGI-2 tests abstract reasoning on grid puzzles that people solve easily and AI systems find hard: infer a hidden rule from a few examples and apply it to a new grid. Made by ARC Prize Foundation (François Chollet, Mike Knoop, Greg Kamradt and co-authors), May 2025.
§ 01 / LEADERBOARD
Top models on ARC-AGI-2, November 2026
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra OpenAI | ARC Prize verified, best result | $10 / $50 | 1.05M | Official leaderboard ARC Prize Foundation |
| 2 | GPT-6.1 Sol OpenAI | ARC Prize verified, best result | $2 / $10 | 1.05M | Official leaderboard ARC Prize Foundation |
| 3 | Claude Opus 5.5 Anthropic | ARC Prize verified, best result | $4 / $20 | 1M | Official leaderboard ARC Prize Foundation |
| 4 | GPT-5.6 Sol OpenAI | ARC Prize verified, best result | $4 / $20 | 1.05M | Official leaderboard ARC Prize Foundation |
| 5 | Gemini 4 Argon Google DeepMind | ARC Prize verified, best result | $2 / $10 | 1M | Official leaderboard ARC Prize Foundation |
| 6 | Claude Opus 5 Anthropic | ARC Prize verified, best result | $5 / $25 | 1M | Official leaderboard ARC Prize Foundation |
| 7 | Claude Fable 5.1 Anthropic | ARC Prize verified, best result | $10 / $50 | 1M | Official leaderboard ARC Prize Foundation |
| 8 | GPT-6 Sol OpenAI | ARC Prize verified, best result | $2 / $10 | 1.05M | Official leaderboard ARC Prize Foundation |
| 9 | Claude Fable 5 Anthropic | ARC Prize verified, best result | $10 / $50 | 1M | Official leaderboard ARC Prize Foundation |
| 9 | Gemini 3.8 Flash Google DeepMind | ARC Prize verified, best result | $0.75 / $3.75 | 1M | Official leaderboard ARC Prize Foundation |
| 11 | GPT-5.5 OpenAI | ARC Prize verified, best result | $5 / $30 | 1M | Official leaderboard ARC Prize Foundation |
| 12 | Gemini 3 Deep Think (2/26) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 12 | Gemini 3.7 Flash Deprecated Google DeepMind | ARC Prize verified, best result | $0.75 / $3.75 | 1M | Official leaderboard ARC Prize Foundation |
| 12 | GPT-5.5 Pro OpenAI | ARC Prize verified, best result | $30 / $180 | 1.05M | Official leaderboard ARC Prize Foundation |
| 15 | Claude Sonnet 5.5 Anthropic | ARC Prize verified, best result | $2 / $10 | 1M | Official leaderboard ARC Prize Foundation |
| 15 | GPT-5.6 Terra OpenAI | ARC Prize verified, best result | $2 / $12 | 1.05M | Official leaderboard ARC Prize Foundation |
| 17 | GPT-5.4 Pro OpenAI | ARC Prize verified, best result | $30 / $180 | 1.05M | Official leaderboard ARC Prize Foundation |
| 18 | Inkling 2 Small Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 19 | DeepSeek V4.1 Flash DeepSeek · listed as “DeepSeek V4.1 Flash 1017” | ARC Prize verified, best result | $0.3 / $1.2 | 1M | Official leaderboard ARC Prize Foundation |
| 20 | Gemini 3.1 Pro Preview Preview Google | ARC Prize verified, best result | $2 / $12 | 1M | Official leaderboard ARC Prize Foundation |
| 21 | Dots3-Note Preview Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 22 | Claude Opus 4.7 Anthropic · listed as “Claude 4.7” | ARC Prize verified, best result | $5 / $25 | 1M | Official leaderboard ARC Prize Foundation |
| 23 | GPT-5.4 OpenAI | ARC Prize verified, best result | $2.5 / $15 | 1.05M | Official leaderboard ARC Prize Foundation |
| 24 | GPT-5.2 (Refinement) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 25 | Claude Opus 4.8 Anthropic | ARC Prize verified, best result | $5 / $25 | 1M | Official leaderboard ARC Prize Foundation |
| 25 | Gemini 3.5 Flash Deprecated Google | ARC Prize verified, best result | $1.5 / $9 | 1M | Official leaderboard ARC Prize Foundation |
| 27 | Claude Opus 4.6 Anthropic · listed as “Claude Opus 4.6 (120K)” | ARC Prize verified, best result | $5 / $25 | 1M | Official leaderboard ARC Prize Foundation |
| 28 | Grok 4.6 SpaceXSI | ARC Prize verified, best result | $2 / $6 | 500K | Official leaderboard ARC Prize Foundation |
| 29 | GLM-5.3-Flash Zhipu (Z.ai / GLM) | ARC Prize verified, best result | $0.15 / $0.5 | 1M | Official leaderboard ARC Prize Foundation |
| 30 | Grok 4.20 (0309) Reasoning SpaceXSI · listed as “Grok 4.20 (Reasoning)” | ARC Prize verified, best result | $1.25 / $2.5 | 1M | Official leaderboard ARC Prize Foundation |
| 31 | DeepSeek V4 Flash Deprecated DeepSeek · listed as “DeepSeek V4 Flash 0731” | ARC Prize verified, best result | $0.44 / $1.32 | 1M | Official leaderboard ARC Prize Foundation |
| 31 | Grok 4.7 SpaceXSI | ARC Prize verified, best result | $2 / $6 | 500K | Official leaderboard ARC Prize Foundation |
| 33 | Claude Haiku 5.5 Anthropic | ARC Prize verified, best result | $0.1 / $0.5 | 1M | Official leaderboard ARC Prize Foundation |
| 33 | DeepSeek V4 Pro DeepSeek · listed as “DeepSeek V4 Pro 0813” | ARC Prize verified, best result | $1.32 / $3.96 | 1M | Official leaderboard ARC Prize Foundation |
| 35 | Claude Sonnet 4.6 Anthropic | ARC Prize verified, best result | $3 / $15 | 1M | Official leaderboard ARC Prize Foundation |
| 35 | Gemini 3.6 Flash Google | ARC Prize verified, best result | $0.75 / $3.75 | 1M | Official leaderboard ARC Prize Foundation |
| 35 | Kimi K3 Moonshot (Kimi) | ARC Prize verified, best result | $3 / $15 | 1M | Official leaderboard ARC Prize Foundation |
| 38 | GPT-5.6 Luna OpenAI · listed as “GPT-5.6 Luna 2026-07-30” | ARC Prize verified, best result | $0.2 / $1.2 | 1.05M | Official leaderboard ARC Prize Foundation |
| 39 | GPT-6 Luna OpenAI | ARC Prize verified, best result | $0.1 / $0.5 | 1.05M | Official leaderboard ARC Prize Foundation |
| 40 | Muse Spark Open Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 41 | GPT-5.2 Pro Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 42 | Gemini 3 Pro (Refinement) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 43 | GPT-5.2 Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 44 | Grok 4.5 SpaceXSI | ARC Prize verified, best result | $2 / $6 | 500K | Official leaderboard ARC Prize Foundation |
| 45 | Gemini 3 Deep Think (Preview) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 46 | Qwen3.8-27B Alibaba (Qwen) | ARC Prize verified, best result | $0.5 / $3 | 1M | Official leaderboard ARC Prize Foundation |
| 47 | Inkling Small Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 48 | Claude Opus 4.5 Anthropic · listed as “Opus 4.5” | ARC Prize verified, best result | $5 / $25 | 200K | Official leaderboard ARC Prize Foundation |
| 49 | Inkling Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 50 | Gemini 3 Flash Preview Preview Google | ARC Prize verified, best result | $0.5 / $3 | 1M | Official leaderboard ARC Prize Foundation |
| 51 | Gemini 3 Pro Preview Retired Google · listed as “Gemini 3 Pro” | ARC Prize verified, best result | $2 / $12 | 1M | Official leaderboard ARC Prize Foundation |
| 52 | Grok 4 (Refinement) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 53 | NVARC Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 54 | Beam 1.0 Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 55 | GLM-5.2 Zhipu (Z.ai / GLM) | ARC Prize verified, best result | $1.4 / $4.4 | 1M | Official leaderboard ARC Prize Foundation |
| 56 | GPT-5.4 mini OpenAI | ARC Prize verified, best result | $0.75 / $4.5 | 400K | Official leaderboard ARC Prize Foundation |
| 57 | GPT-5 Pro Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 58 | GPT-5.1 Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 59 | Grok 4 Retired SpaceXSI · listed as “Grok 4 (Thinking)” | ARC Prize verified, best result | $3 / $15 | 256K | Official leaderboard ARC Prize Foundation |
| 60 | Claude Sonnet 4.5 Deprecated Anthropic | ARC Prize verified, best result | $3 / $15 | 200K | Official leaderboard ARC Prize Foundation |
| 61 | Kimi K2.5 Retired Moonshot (Kimi) | ARC Prize verified, best result | $0.6 / $3 | 262K | Official leaderboard ARC Prize Foundation |
| 62 | Gemini 3.5 Flash-Lite Google | ARC Prize verified, best result | $0.3 / $2.5 | 1M | Official leaderboard ARC Prize Foundation |
| 63 | GPT-5 Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 64 | Claude Opus 4 Retired Anthropic | ARC Prize verified, best result | $15 / $75 | 200K | Official leaderboard ARC Prize Foundation |
| 65 | o3 Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 66 | Tiny Recursion Model (TRM) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 67 | o4-mini Retired OpenAI | ARC Prize verified, best result | $1.1 / $4.4 | 200K | Official leaderboard ARC Prize Foundation |
| 68 | Claude Sonnet 4 Retired Anthropic | ARC Prize verified, best result | $3 / $15 | 200K | Official leaderboard ARC Prize Foundation |
| 69 | GPT-5.4 nano Deprecated OpenAI | ARC Prize verified, best result | $0.2 / $1.25 | 400K | Official leaderboard ARC Prize Foundation |
| 70 | Grok 4 Fast Retired SpaceXSI · listed as “Grok 4 (Fast Reasoning)” | ARC Prize verified, best result | $0.2 / $0.5 | 2M | Official leaderboard ARC Prize Foundation |
| 71 | Gemini 2.5 Pro Deprecated Google | ARC Prize verified, best result | $1.25 / $10 | 1M | Official leaderboard ARC Prize Foundation |
| 71 | GLM-5 Zhipu (Z.ai / GLM) | ARC Prize verified, best result | $1 / $3.2 | 200K | Official leaderboard ARC Prize Foundation |
| 71 | MiniMax M2.5 Deprecated MiniMax | ARC Prize verified, best result | $0.3 / $1.2 | 205K | Official leaderboard ARC Prize Foundation |
| 71 | o3-Pro Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 75 | GPT-5 Mini Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 76 | Claude Haiku 4.5 Anthropic | ARC Prize verified, best result | $1 / $5 | 200K | Official leaderboard ARC Prize Foundation |
| 76 | Deepseek V3.2 Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 76 | o3 (Preview) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 79 | o3-mini Retired OpenAI | ARC Prize verified, best result | $1.1 / $4.4 | 200K | Official leaderboard ARC Prize Foundation |
| 80 | GPT-5 Nano Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 81 | ARChitects Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 81 | Gemini 2.5 Flash Deprecated Google · listed as “Gemini 2.5 Flash (Preview)” | ARC Prize verified, best result | $0.3 / $2.5 | 1M | Official leaderboard ARC Prize Foundation |
| 83 | Hierarchical Reasoning Model (HRM) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 84 | Nemotron 3.5 Super VL 120B A12B NVIDIA · listed as “Nemotron 3.5 Super VL” | ARC Prize verified, best result | — | 1M | Official leaderboard ARC Prize Foundation |
| 85 | Icecuber Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 86 | Codex Mini (Latest) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 86 | DeepSeek R1 Retired DeepSeek | ARC Prize verified, best result | $0.55 / $2.19 | 64K | Official leaderboard ARC Prize Foundation |
| 86 | Gemini 2.0 Flash Retired Google | ARC Prize verified, best result | $0.1 / $0.4 | 1M | Official leaderboard ARC Prize Foundation |
| 86 | Qwen3 235B A22B Instruct 2507 Retired Alibaba (Qwen) | ARC Prize verified, best result | $0.23 / $0.92 | 262K | Official leaderboard ARC Prize Foundation |
| 90 | Deepseek R1 (05/28) Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 91 | Claude Sonnet 3.7 Retired Anthropic · listed as “Claude 3.7” | ARC Prize verified, best result | $3 / $15 | 200K | Official leaderboard ARC Prize Foundation |
| 92 | Gemini 1.5 Pro Retired Google | ARC Prize verified, best result | $1.25 / $5 | 2M | Official leaderboard ARC Prize Foundation |
| 92 | GPT-4.5 preview Retired OpenAI · listed as “GPT-4.5” | ARC Prize verified, best result | $75 / $150 | 128K | Official leaderboard ARC Prize Foundation |
| 92 | o1-mini Retired OpenAI | ARC Prize verified, best result | $3 / $12 | 128K | Official leaderboard ARC Prize Foundation |
| 95 | GPT-4.1 Deprecated OpenAI | ARC Prize verified, best result | $2 / $8 | 1M | Official leaderboard ARC Prize Foundation |
| 95 | Grok 3 Mini Retired SpaceXSI | ARC Prize verified, best result | $0.3 / $0.5 | 131K | Official leaderboard ARC Prize Foundation |
| 97 | GPT-4.1 mini Deprecated OpenAI | ARC Prize verified, best result | $0.4 / $1.6 | 1M | Official leaderboard ARC Prize Foundation |
| 97 | GPT-4.1-Nano Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 97 | GPT-4o Deprecated OpenAI | ARC Prize verified, best result | $2.5 / $10 | 128K | Official leaderboard ARC Prize Foundation |
| 97 | GPT-4o mini Deprecated OpenAI | ARC Prize verified, best result | $0.15 / $0.6 | 128K | Official leaderboard ARC Prize Foundation |
| 97 | Grok 3 Retired SpaceXSI | ARC Prize verified, best result | $3 / $15 | 131K | Official leaderboard ARC Prize Foundation |
| 97 | Llama 4 Maverick Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 97 | Llama 4 Scout Not in our price index | ARC Prize verified, best result | — | — | Official leaderboard ARC Prize Foundation |
| 97 | Magistral Medium Retired Mistral | ARC Prize verified, best result | $2 / $5 | — | Official leaderboard ARC Prize Foundation |
| 97 | Magistral Small Retired Mistral | ARC Prize verified, best result | $0.5 / $1.5 | — | Official leaderboard ARC Prize Foundation |
§ 02 / SCORE VS PRICE
ARC-AGI-2: score against price
§ 03 / PROVENANCE
Can you trust ARC-AGI-2 scores?
105 scores on this page come from the benchmark's own leaderboard (ARC Prize Foundation), where every model runs on the same harness. The other 4 are labs' reports on their own models. 4 models have numbers from two sources for the same setting, and every pair agrees within 1 point.
- ARC Prize verifies scores on its semi-private set. That set has been exposed to API providers during testing, so ARC Prize itself says limited leakage is possible.
- Accuracy alone leaves out cost: ARC Prize reports both, and the same model can score very differently at different reasoning budgets.
- Labs that report their own ARC-AGI-2 scores may use the public set, where puzzles are openly available.
§ 04 / HOW IT WORKS
How ARC-AGI-2 works
The task
Each task shows a few input and output grids that demonstrate a hidden transformation. The model has to produce the output grid for a new input. Version 2 adds harder puzzles that need symbolic interpretation and rules that combine or depend on context.
The score
The share of evaluation tasks solved with up to two attempts per task (pass@2). ARC Prize publishes the cost per task beside every score.
Test set
120 tasks in each evaluation set (public, semi-private, private), as published by the authors.
Made by
ARC Prize Foundation (François Chollet, Mike Knoop, Greg Kamradt and co-authors), first published May 2025.
Official sources
- arcprize.org/arc-agi/2
- Paper: arxiv.org/abs/2505.11831
- Definition read on 2026-11-08
§ 05 / SOURCES
Sources and freshness
109 published scores from 3 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.
- Official leaderboard ARC Prize Foundation
- Vendor-reported Google DeepMind
- Vendor-reported OpenAI
ARC-AGI-2 FAQ
Answers derived from the 109 published scores on this page and the current price index.
Q · 01 Which AI model has the highest ARC-AGI-2 score?
+
GPT-6 Astra with 95% (ARC Prize verified, best result), on ARC Prize Foundation's own leaderboard on September 2, 2026. Next: GPT-6.1 Sol at 94.2% and Claude Opus 5.5 at 93.3%.
Q · 02 What is the cheapest model near the top of ARC-AGI-2?
+
Gemini 4 Argon scores 91.3%, 3.7 points behind the leader, at $1.21 per 1M tokens on our effective blend ($2 input, $10 output). GPT-6 Astra costs $6.82 on the same blend.
Q · 03 Are these ARC-AGI-2 scores independently verified?
+
105 of 109 published scores come from ARC Prize Foundation; the other 4 are vendor reports. Each row carries a badge that says which, and a link to the source.
Q · 04 What does ARC-AGI-2 measure?
+
ARC-AGI-2 tests abstract reasoning on grid puzzles that people solve easily and AI systems find hard: infer a hidden rule from a few examples and apply it to a new grid. Each task shows a few input and output grids that demonstrate a hidden transformation. The model has to produce the output grid for a new input. Version 2 adds harder puzzles that need symbolic interpretation and rules that combine or depend on context.
Q · 05 How is the ARC-AGI-2 score calculated?
+
The share of evaluation tasks solved with up to two attempts per task (pass@2). ARC Prize publishes the cost per task beside every score. The test set has 120 tasks in each evaluation set (public, semi-private, private).
Q · 06 How current is this ARC-AGI-2 leaderboard?
+
The newest score was published on November 5, 2026. Every row shows the date its source published it, and we last read all of them on November 8, 2026. Prices come from our model price index, where each rate carries the date it was read on the vendor's page.