ARC-AGI-3
Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions.
Scores depend on the harness. ARC Prize reports GPT-6 Astra at 62.7 percent with its Standard harness and 99.9 percent with a Provider Adapter harness. A six-game preview ran from July 2025.
- Made by
- ARC Prize Foundation
- Released
- March 2026
- Kind
- Puzzles and chess
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- Public, plus a private test set
- Strongest model tested
- Claude Fable 5.2 #1 on the capability index
- Last in a launch post
- September 2026
- Last model added
- November 2026
- Scores refreshed
- November 12th, 2026
In the authors’ words
We introduce ARC-AGI-3, an interactive benchmark for studying agentic intelligence through novel, abstract, turn-based environments in which agents must explore, infer goals, build internal models of environment dynamics, and plan effective action sequences without explicit instructions. Like its predecessors ARC-AGI-1 and 2, ARC-AGI-3 focuses entirely on evaluating fluid adaptive efficiency on novel tasks, while avoiding language and external knowledge. ARC-AGI-3 environments only leverage Core Knowledge priors and are difficulty-calibrated via extensive testing with human test-takers. Our testing shows humans can solve 100% of the environments, in contrast to frontier AI systems which, as of March 2026, score below 1%. In this paper, we present the benchmark design, its efficiency-based scoring framework grounded in human action baselines, and the methodology used to construct, validate, and calibrate the environments.
27
Models scored on its own leaderboard
5
Launch posts list it, from 2 labs
2
System cards report it, from 1 lab
GPT-6 Astra leads with 99.95%
Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
GPT-6 Astra leads with 99.95%, claims included
Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.
- 99.95 GPT-6 Astra
- 99.1
Gemini 4 Carbon - 97.7 Claude Fable 5.2
- 97.5 Gemini 4 Argon
- 96.4 GPT-6.1 Sol
- 72.4 Claude Sonnet 5.5
- 35.0 Gemini 3.8 Flash
- 30.2 Claude Opus 5
- 23.0 GPT-6 Sol
- 10.1 Grok 4.7
- 7.78 GPT-5.6 Sol
- 5.81
DeepSeek V4.1 Flash 1017 - 3.26 DeepSeek V4.1 Flash
- 2.82 Claude Haiku 5.5
- 2.11 Grok 4.6
- 1.52 Claude Opus 4.8
- 0.80 GPT-5.6 Terra
- 0.59 GPT-6 Luna
- 0.51 Claude Opus 4.6
- 0.43 GPT-5.5
- 0.42 Gemini 3.1 Pro
- 0.42
Muse Spark Open - 0.32 Grok 4.5
- 0.21 GPT-5.4
- 0.18 Claude Opus 4.7
- 0.18 GPT-5.6 Luna
- 0.09 Grok 4.20
- 1 GPT-6 Astra high 99.95%
- 2 Gemini 4 Carbon medium 99.1%
- 3 Claude Fable 5.2 high 97.7%
- 4 Gemini 4 Argon low 97.5%
- 5 GPT-6.1 Sol xhigh 96.4%
- 6 Claude Sonnet 5.5 xhigh 72.4%
- 7 Gemini 3.8 Flash high 35%
- 8 Claude Opus 5 high 30.2%
- 9 GPT-6 Sol max 23%
- 10 Grok 4.7 high 10.1%
- 11 GPT-5.6 Sol max 7.78%
- 12 DeepSeek V4.1 Flash 1017 max 5.81%
- 13 DeepSeek V4.1 Flash low 3.26%
- 14 Claude Haiku 5.5 max 2.82%
- 15 Grok 4.6 xhigh 2.11%
- 16 Claude Opus 4.8 high 1.52%
- 17 GPT-5.6 Terra max 0.80%
- 18 GPT-6 Luna max 0.59%
- 19 Claude Opus 4.6 max 0.51%
- 20 GPT-5.5 high 0.43%
- 21 Gemini 3.1 Pro 0.42%
- 22 Muse Spark Open xhigh 0.42%
- 23 Grok 4.5 medium 0.32%
- 24 GPT-5.4 high 0.21%
- 25 Claude Opus 4.7 high 0.18%
- 26 GPT-5.6 Luna max 0.18%
- 27 Grok 4.20 beta 0.09%
Source: Official ARC-AGI-3 leaderboard
- 99.95 GPT-6 Astra
- 99.1 Gemini 4 Carbon
- 97.7 Claude Fable 5.2
- 97.5 Gemini 4 Argon
- 96.4 GPT-6.1 Sol
- 72.4 Claude Sonnet 5.5
- 35.0 Gemini 3.8 Flash
- 30.2 Claude Opus 5
- 23.0 GPT-6 Sol
- 10.1 Grok 4.7
- 7.78 GPT-5.6 Sol
- 5.81 DeepSeek V4.1 Flash 1017
- 3.26 DeepSeek V4.1 Flash
- 2.82 Claude Haiku 5.5
- 2.11 Grok 4.6
- 1.52 Claude Opus 4.8
- 0.80 GPT-5.6 Terra
- 0.59 GPT-6 Luna
- 0.51 Claude Opus 4.6
- 0.43 GPT-5.5
- 0.42 Gemini 3.1 Pro
- 0.42 Muse Spark Open
- 0.32 Grok 4.5
- 0.21 GPT-5.4
- 0.18 Claude Opus 4.7
- 0.18 GPT-5.6 Luna
- 0.09 Grok 4.20
- 1 GPT-6 Astra high 99.95%
- 2 Gemini 4 Carbon medium 99.1%
- 3 Claude Fable 5.2 high 97.7%
- 4 Gemini 4 Argon low 97.5%
- 5 GPT-6.1 Sol xhigh 96.4%
- 6 Claude Sonnet 5.5 xhigh 72.4%
- 7 Gemini 3.8 Flash high 35%
- 8 Claude Opus 5 30.2%
- 9 GPT-6 Sol max 23%
- 10 Grok 4.7 high 10.1%
- 11 GPT-5.6 Sol max 7.78%
- 12 DeepSeek V4.1 Flash 1017 max 5.81%
- 13 DeepSeek V4.1 Flash low 3.26%
- 14 Claude Haiku 5.5 max 2.82%
- 15 Grok 4.6 xhigh 2.11%
- 16 Claude Opus 4.8 high 1.52%
- 17 GPT-5.6 Terra max 0.80%
- 18 GPT-6 Luna max 0.59%
- 19 Claude Opus 4.6 max 0.51%
- 20 GPT-5.5 high 0.43%
- 21 Gemini 3.1 Pro 0.42%
- 22 Muse Spark Open xhigh 0.42%
- 23 Grok 4.5 medium 0.32%
- 24 GPT-5.4 high 0.21%
- 25 Claude Opus 4.7 high 0.18%
- 26 GPT-5.6 Luna max 0.18%
- 27 Grok 4.20 beta 0.09%
Sources: Official ARC-AGI-3 leaderboard and labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 99.95% | high, Provider Adapter | Sep 2026 | Official leaderboard |
| 2 | Gemini 4 Carbon | 99.1% | medium, Provider Adapter | Oct 2026 | Official leaderboard |
| 3 | Claude Fable 5.2 | 97.7% | high, Provider Adapter | Nov 2026 | Official leaderboard |
| 4 | Gemini 4 Argon | 97.5% | low, Provider Adapter | Sep 2026 | Official leaderboard |
| 5 | GPT-6.1 Sol | 96.4% | xhigh, Provider Adapter | Sep 2026 | Official leaderboard |
| 6 | Claude Sonnet 5.5 | 72.4% | xhigh, Provider Adapter | Sep 2026 | Official leaderboard |
| 7 | Gemini 3.8 Flash | 35% | high, Provider Adapter | Sep 2026 | Official leaderboard |
| 8 | Claude Opus 5 | 30.2% | high, standard | Jul 2026 | Official leaderboard |
| 9 | GPT-6 Sol | 23% | max, Provider Adapter | Sep 2026 | Official leaderboard |
| 10 | Grok 4.7 | 10.1% | high, Provider Adapter | Sep 2026 | Official leaderboard |
| 11 | GPT-5.6 Sol | 7.78% | max, standard | Jul 2026 | Official leaderboard |
| 12 | DeepSeek V4.1 Flash 1017 | 5.81% | max, Provider Adapter | Oct 2026 | Official leaderboard |
| 13 | DeepSeek V4.1 Flash | 3.26% | low, Provider Adapter | Sep 2026 | Official leaderboard |
| 14 | Claude Haiku 5.5 | 2.82% | max, Provider Adapter | Oct 2026 | Official leaderboard |
| 15 | Grok 4.6 | 2.11% | xhigh, standard | Aug 2026 | Official leaderboard |
| 16 | Claude Opus 4.8 | 1.52% | high, standard | May 2026 | Official leaderboard |
| 17 | GPT-5.6 Terra | 0.80% | max, standard | Jul 2026 | Official leaderboard |
| 18 | GPT-6 Luna | 0.59% | max, Provider Adapter | Sep 2026 | Official leaderboard |
| 19 | Claude Opus 4.6 | 0.51% | max, standard | Feb 2026 | Official leaderboard |
| 20 | GPT-5.5 | 0.43% | high, standard | Apr 2026 | Official leaderboard |
| 21 | Gemini 3.1 Pro | 0.42% | standard | Feb 2026 | Official leaderboard |
| 22 | Muse Spark Open | 0.42% | xhigh, Provider Adapter | Oct 2026 | Official leaderboard |
| 23 | Grok 4.5 | 0.32% | medium, standard | Jul 2026 | Official leaderboard |
| 24 | GPT-5.4 | 0.21% | high, standard | Mar 2026 | Official leaderboard |
| 25 | Claude Opus 4.7 | 0.18% | high, standard | Apr 2026 | Official leaderboard |
| 26 | GPT-5.6 Luna | 0.18% | max, standard | Jul 2026 | Official leaderboard |
| 27 | Grok 4.20 | 0.09% | beta, standard | Feb 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 99.95% | high, Provider Adapter | Sep 2026 | Official leaderboard |
| 2 | Gemini 4 Carbon | 99.1% | medium, Provider Adapter | Oct 2026 | Official leaderboard |
| 3 | Claude Fable 5.2 | 97.7% | high, Provider Adapter | Nov 2026 | Official leaderboard |
| 4 | Gemini 4 Argon | 97.5% | low, Provider Adapter | Sep 2026 | Official leaderboard |
| 5 | GPT-6.1 Sol | 96.4% | xhigh, Provider Adapter | Sep 2026 | Official leaderboard |
| 6 | Claude Sonnet 5.5 | 72.4% | xhigh, Provider Adapter | Sep 2026 | Official leaderboard |
| 7 | Gemini 3.8 Flash | 35% | high, Provider Adapter | Sep 2026 | Official leaderboard |
| 8 | Claude Opus 5 | 30.2% | Jul 2026 | Lab model card | |
| 9 | GPT-6 Sol | 23% | max, Provider Adapter | Sep 2026 | Official leaderboard |
| 10 | Grok 4.7 | 10.1% | high, Provider Adapter | Sep 2026 | Official leaderboard |
| 11 | GPT-5.6 Sol | 7.78% | max, standard | Jul 2026 | Official leaderboard |
| 12 | DeepSeek V4.1 Flash 1017 | 5.81% | max, Provider Adapter | Oct 2026 | Official leaderboard |
| 13 | DeepSeek V4.1 Flash | 3.26% | low, Provider Adapter | Sep 2026 | Official leaderboard |
| 14 | Claude Haiku 5.5 | 2.82% | max, Provider Adapter | Oct 2026 | Official leaderboard |
| 15 | Grok 4.6 | 2.11% | xhigh, standard | Aug 2026 | Official leaderboard |
| 16 | Claude Opus 4.8 | 1.52% | high, standard | May 2026 | Official leaderboard |
| 17 | GPT-5.6 Terra | 0.80% | max, standard | Jul 2026 | Official leaderboard |
| 18 | GPT-6 Luna | 0.59% | max, Provider Adapter | Sep 2026 | Official leaderboard |
| 19 | Claude Opus 4.6 | 0.51% | max, standard | Feb 2026 | Official leaderboard |
| 20 | GPT-5.5 | 0.43% | high, standard | Apr 2026 | Official leaderboard |
| 21 | Gemini 3.1 Pro | 0.42% | standard | Feb 2026 | Official leaderboard |
| 22 | Muse Spark Open | 0.42% | xhigh, Provider Adapter | Oct 2026 | Official leaderboard |
| 23 | Grok 4.5 | 0.32% | medium, standard | Jul 2026 | Official leaderboard |
| 24 | GPT-5.4 | 0.21% | high, standard | Mar 2026 | Official leaderboard |
| 25 | Claude Opus 4.7 | 0.18% | high, standard | Apr 2026 | Official leaderboard |
| 26 | GPT-5.6 Luna | 0.18% | max, standard | Jul 2026 | Official leaderboard |
| 27 | Grok 4.20 | 0.09% | beta, standard | Feb 2026 | Official leaderboard |
Labs report scores for 6 models in their cards. Switch the scores above to rank them together.
The best score, 99.95%, costs $18,820 for the full run
Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Sol at max comes within 3.8 points of the best for $3,817 for the full run, 5 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Google DeepMind OpenAI DeepSeek Anthropic
Cost frontier
One model at each effort level
Source: Official ARC-AGI-3 leaderboard
| Model | Effort | Harness | Score | Cost of the full run | Output tokens |
|---|---|---|---|---|---|
| GPT-6 Astra | high | Provider Adapter | 99.95% | $18,820 | |
| Gemini 4 Carbon | medium | Provider Adapter | 99.1% | $4,560 | |
| GPT-6 Astra | max | Provider Adapter | 98.5% | $17,330 | |
| GPT-6 Astra | xhigh | Provider Adapter | 98.4% | $18,150 | |
| GPT-6 Astra | medium | Provider Adapter | 98.4% | $19,280 | |
| GPT-6 Astra | low | Provider Adapter | 98% | $21,300 | |
| Claude Fable 5.2 | high | Provider Adapter | 97.7% | $18,620 | |
| Gemini 4 Carbon | low | Provider Adapter | 97.7% | $4,351 | |
| Gemini 4 Argon | low | Provider Adapter | 97.5% | $4,770 | |
| Gemini 4 Argon | medium | Provider Adapter | 97.3% | $4,981 | |
| Gemini 4 Argon | high | Provider Adapter | 97.3% | $4,626 | |
| Claude Fable 5.2 | xhigh | Provider Adapter | 97.3% | $18,250 | |
| Gemini 4 Carbon | high | Provider Adapter | 96.8% | $4,350 | |
| Claude Fable 5.2 | medium | Provider Adapter | 96.8% | $21,660 | |
| GPT-6 Astra | none | Provider Adapter | 96.7% | $23,460 | |
| GPT-6.1 Sol | xhigh | Provider Adapter | 96.4% | $4,360 | |
| GPT-6.1 Sol | max | Provider Adapter | 96.2% | $3,817 | |
| GPT-6.1 Sol | high | Provider Adapter | 95% | $5,404 | |
| Claude Fable 5.2 | max | Provider Adapter | 94.9% | $17,880 | |
| Claude Fable 5.2 | low | Provider Adapter | 92.3% | $27,380 | |
| GPT-6.1 Sol | medium | Provider Adapter | 91% | $5,835 | |
| GPT-6.1 Sol | low | Provider Adapter | 82.8% | $6,622 | |
| Claude Fable 5.2 | xhigh | standard | 81.2% | $21,950 | |
| Claude Fable 5.2 | high | standard | 79.3% | $25,170 | |
| Claude Fable 5.2 | medium | standard | 73.3% | $29,030 | |
| Claude Sonnet 5.5 | xhigh | Provider Adapter | 72.4% | $8,170 | |
| Claude Fable 5.2 | max | standard | 65.7% | $17,140 | |
| Gemini 4 Carbon | low | standard | 65.1% | $7,244 | |
| Gemini 4 Carbon | medium | standard | 63.9% | $7,180 | |
| GPT-6 Astra | max | standard | 62.7% | $26,100 | |
| GPT-6 Astra | xhigh | standard | 59.3% | $37,320 | |
| GPT-6 Astra | high | standard | 54.8% | $40,700 | |
| GPT-6.1 Sol | max | standard | 52.7% | $7,632 | |
| Claude Fable 5.2 | low | standard | 52% | $34,910 | |
| Gemini 4 Argon | low | standard | 49.8% | $5,821 | |
| Gemini 4 Argon | medium | standard | 46.3% | $6,179 | |
| GPT-6.1 Sol | xhigh | standard | 39.9% | $9,104 | |
| GPT-6 Astra | medium | standard | 38.6% | $48,090 | |
| Claude Sonnet 5.5 | high | Provider Adapter | 35.5% | $10,240 | |
| GPT-6 Astra | none | standard | 35.2% | $49,790 | |
| Gemini 3.8 Flash | high | Provider Adapter | 35% | $4,522 | |
| Claude Opus 5 | high | standard | 30.2% | $20,660 | |
| GPT-6.1 Sol | high | standard | 26.7% | $9,758 | |
| Claude Sonnet 5.5 | xhigh | standard | 24.3% | $7,394 | |
| Gemini 3.8 Flash | medium | Provider Adapter | 24.2% | $3,826 | |
| GPT-6 Sol | max | Provider Adapter | 23% | $8,722 | |
| GPT-6 Astra | low | standard | 17.4% | $38,170 | |
| Gemini 3.8 Flash | low | Provider Adapter | 15.1% | $3,158 | |
| GPT-6.1 Sol | medium | standard | 10.6% | $7,370 | |
| Gemini 3.8 Flash | high | standard | 10.4% | $4,398 | |
| Grok 4.7 | high | Provider Adapter | 10.1% | $4,774 | |
| GPT-6 Sol | xhigh | Provider Adapter | 9.45% | $6,907 | |
| Grok 4.7 | medium | Provider Adapter | 8.37% | $5,060 | |
| GPT-5.6 Sol | max | standard | 7.78% | $25,060 | |
| Grok 4.7 | xhigh | Provider Adapter | 7.02% | $4,576 | |
| GPT-5.6 Sol | xhigh | standard | 6.99% | $19,220 | |
| Gemini 3.8 Flash | low | standard | 5.99% | $2,669 | |
| GPT-6 Sol | high | Provider Adapter | 5.86% | $5,908 | |
| DeepSeek V4.1 Flash 1017 | max | Provider Adapter | 5.81% | $1,179 | |
| GPT-6 Sol | medium | Provider Adapter | 4.93% | $5,889 | |
| Claude Sonnet 5.5 | medium | Provider Adapter | 4.89% | $5,250 | |
| GPT-6 Sol | max | standard | 4.62% | $5,554 | |
| Gemini 3.8 Flash | medium | standard | 3.99% | $2,728 | |
| GPT-6.1 Sol | low | standard | 3.92% | $7,111 | |
| DeepSeek V4.1 Flash | low | Provider Adapter | 3.26% | $878 | |
| DeepSeek V4.1 Flash | high | Provider Adapter | 3.15% | $1,022 | |
| Claude Haiku 5.5 | max | Provider Adapter | 2.82% | $437 | |
| DeepSeek V4.1 Flash | max | Provider Adapter | 2.68% | $1,309 | |
| GPT-5.6 Sol | high | standard | 2.15% | $15,180 | |
| Grok 4.6 | xhigh | standard | 2.11% | $5,612 | |
| Grok 4.7 | xhigh | standard | 1.82% | $2,732 | |
| GPT-6 Sol | xhigh | standard | 1.80% | $5,106 | |
| GPT-6 Sol | low | Provider Adapter | 1.76% | $4,745 | |
| Grok 4.7 | high | standard | 1.73% | $3,742 | |
| Grok 4.7 | medium | standard | 1.71% | $3,795 | |
| Claude Opus 4.8 | high | standard | 1.52% | $10,000 | |
| GPT-6 Sol | none | Provider Adapter | 1.21% | $4,234 | |
| Claude Sonnet 5.5 | high | standard | 1.20% | $5,487 | |
| GPT-5.6 Sol | medium | standard | 1.07% | $12,970 | |
| GPT-6 Sol | high | standard | 0.90% | $5,072 | |
| DeepSeek V4.1 Flash | max | standard | 0.85% | $1,165 | |
| GPT-5.6 Terra | max | standard | 0.80% | $7,918 | |
| Grok 4.7 | low | Provider Adapter | 0.77% | $5,754 | |
| GPT-5.6 Terra | xhigh | standard | 0.65% | $6,804 | |
| Claude Sonnet 5.5 | low | Provider Adapter | 0.62% | $4,797 | |
| GPT-6 Luna | max | Provider Adapter | 0.59% | $237 | |
| Claude Opus 4.6 | max | standard | 0.51% | $8,866 | |
| GPT-6 Luna | xhigh | Provider Adapter | 0.51% | $223 | |
| GPT-5.6 Terra | high | standard | 0.49% | $5,881 | |
| DeepSeek V4.1 Flash | high | standard | 0.48% | $1,006 | |
| DeepSeek V4.1 Flash | none | Provider Adapter | 0.48% | $648 | |
| Claude Haiku 5.5 | max | standard | 0.45% | $388 | |
| GPT-5.5 | high | standard | 0.43% | $10,000 | |
| Gemini 3.1 Pro | standard | 0.42% | $2,214 | ||
| GPT-6 Sol | medium | standard | 0.42% | $4,864 | |
| Muse Spark Open | xhigh | Provider Adapter | 0.42% | $2,996 | |
| Claude Sonnet 5.5 | low | standard | 0.40% | $4,757 | |
| Claude Sonnet 5.5 | medium | standard | 0.40% | $4,840 | |
| GPT-6 Luna | high | Provider Adapter | 0.38% | $224 | |
| DeepSeek V4.1 Flash | low | standard | 0.35% | $821 | |
| DeepSeek V4.1 Flash | none | standard | 0.34% | $735 | |
| GPT-5.6 Sol | low | standard | 0.33% | $12,810 | |
| Grok 4.5 | medium | standard | 0.32% | $8,458 | |
| GPT-6 Luna | medium | Provider Adapter | 0.32% | $214 | |
| Grok 4.5 | high | standard | 0.30% | $6,893 | |
| GPT-6 Sol | none | standard | 0.28% | $4,656 | |
| Grok 4.5 | low | standard | 0.26% | $8,718 | |
| Grok 4.7 | low | standard | 0.25% | $4,815 | |
| GPT-6 Luna | low | Provider Adapter | 0.24% | $214 | |
| DeepSeek V4.1 Flash 1017 | max | standard | 0.22% | $1,004 | |
| GPT-5.4 | high | standard | 0.21% | $5,187 | |
| GPT-6 Luna | medium | standard | 0.19% | $241 | |
| GPT-6 Luna | high | standard | 0.18% | $256 | |
| Claude Opus 4.7 | high | standard | 0.18% | $10,000 | |
| GPT-5.6 Luna | max | standard | 0.18% | $3,189 | |
| GPT-5.6 Luna | low | standard | 0.17% | $2,307 | |
| GPT-5.6 Luna | medium | standard | 0.17% | $2,379 | |
| GPT-6 Luna | xhigh | standard | 0.16% | $263 | |
| Muse Spark Open | xhigh | standard | 0.14% | $3,908 | |
| GPT-6 Sol | low | standard | 0.13% | $4,752 | |
| GPT-6 Luna | max | standard | 0.10% | $262 | |
| GPT-5.6 Luna | high | standard | 0.10% | $2,482 | |
| Grok 4.20 | beta | standard | 0.09% | $3,775 | |
| GPT-5.6 Terra | medium | standard | 0.08% | $5,895 | |
| GPT-6 Luna | none | Provider Adapter | 0.05% | $205 | |
| GPT-6 Luna | low | standard | 0.03% | $229 | |
| GPT-6 Luna | none | standard | 0.03% | $230 | |
| GPT-5.6 Luna | xhigh | standard | 0.02% | $2,802 | |
| GPT-5.6 Terra | low | standard | 0.01% | $5,068 |
6 models have held number one on the official leaderboard since its release in March 2026 6 models have held number one with claims included since its release in March 2026
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.
Anthropic OpenAI Google DeepMind
Sources: Official ARC-AGI-3 leaderboard and labs’ model cards
GPT-6 Astra on the official leaderboard passed 85% in September 2026
All the scores we have, one dot each, so a model tested at several settings appears more than once.
On the benchmark’s own leaderboard Reported by the lab in its model card
Best so far
Sources: Official ARC-AGI-3 leaderboard and labs’ model cards
2 labs put it in a launch post in the last six months
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Sep 3, 2026 | GPT-6 Astra | 99.9 | ARC-AGI-3 (responses API harness) |
| Jul 24, 2026 | Claude Opus 5 | 30.2 | Novel problem-solving ARC-AGI-3 (high) |
| Jul 9, 2026 | GPT-5.6 Luna | 0.18 | ARC-AGI-3 |
| Jul 9, 2026 | GPT-5.6 Terra | 0.8 | ARC-AGI-3 |
| Jul 9, 2026 | GPT-5.6 Sol | 7.78 | ARC-AGI-3 |
2 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 9, 2026 | Claude Fable 5.2 system card | 79.3 | ARC-AGI-3, verified headline Standard harness score, High effort. Provider Adapter 97.72% at High. At Max: Standard 65.66%, Provider Adapter 94.90% (Section 8.15) | Capabilities |
| Jul 24, 2026 | Claude Opus 5 system card | 30.2 | ARC-AGI-3 semi-private set, RHAE score 30.16% at high effort, max effort not available at release, verified by ARC Prize Foundation (Table 8.1.A, 8.14.2) | Capabilities |
6 lab claims next to the leaderboard
Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 5 of them match, which usually means one side copied the other. Every benchmark.
| Model | Lab says | Board says | Gap |
|---|---|---|---|
| Claude Fable 5.2 | 79.3% ARC-AGI-3, verified headline Standard harness score, High effort. Provider Adapter 97.72% at High. At Max: Standard 65.66%, Provider Adapter 94.90% (Section 8.15) | 97.7% Leaderboard, high, Provider Adapter | −18 pts |
| GPT-6 Astra | 99.9% ARC-AGI-3 (responses API harness) | 99.95% Leaderboard, high, Provider Adapter | Same |
| Claude Opus 5 | 30.2% Novel problem-solving ARC-AGI-3 (high) | 30.2% Leaderboard, high, standard | Same |
| GPT-5.6 Sol | 7.78% ARC-AGI-3 | 7.78% Leaderboard, max, standard | Same |
| GPT-5.6 Terra | 0.80% ARC-AGI-3 | 0.80% Leaderboard, max, standard | Same |
| GPT-5.6 Luna | 0.18% ARC-AGI-3 | 0.18% Leaderboard, max, standard | Same |
Versions
We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.
OpenAI DeepSeek xAI Google DeepMind Anthropic
Source: Official leaderboards
More in Games and long tasks
- Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Each game scores in-game progress or outcomes, and the board ranks models by their average rank across the games. Gemini 2.5 Pro 3.90
- MineDojo Fading Agents complete open-ended tasks in Minecraft, such as mining blocks and fighting mobs, from a large simulation suite. Programmatic checks score many tasks.
Seed2.0 Pro 49% - TextQuests Only in system cards Agents play 25 classic interactive fiction games. The score is harmful actions per 100 steps, where lower is better. The board also reports game progress. Llama 4 Scout 7.00
- GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. GPT-6.1 Terra 76.8%
- Swarm Bench Only in system cards Models orchestrate agent swarms on complex tasks that benefit from coordinated decomposition and parallel execution. Kimi K3 76.3%
- Anthropic large agent teams tasks Only in system cards Internal 24-hour tasks for teams of 1-100 agents: build a company knowledge base, or formalize a theorem in Lean.