Skip to content

Benchmark

ARC-AGI-1

Scores are percentages. Higher is better.

Models measured
16
Scores recorded
26
Sources
1
IQ weight
2.37

Score distribution

90.8%
Mean
94.5%
Median
11.2%
Std dev
86.2%
10th percentile
98.5%
90th percentile
54.8%
Min
54.8%98.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.37 for chat models, measured over 16 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Fable 5OpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.98.5%
2Claude Fable 5System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.98.5%
3GPT-6 AstraOpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.98.5%
4Claude Fable 5.1OpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.97.5%
5Claude Fable 5.1System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.97.5%
6Claude Opus 5OpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.97.5%
7Claude Opus 5Opus 5 System Card capability summary · adaptive thinking · max effort unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.97.5%
8Claude Opus 5System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.97.5%
9GPT-5.6 SolOpenAI launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.97.5%
10GPT-5.6 Solxhigh· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.97.5%
11GPT-5.6 SolSystem Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.96.5%
12GPT 5.5Verified · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.95%
13GPT-5.4 ProVerified· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.94.5%
14Claude Opus 4.6high effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.94%
15GPT 5.4Verified· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.93.7%
16Claude Opus 4.7private validation · high effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.93.5%
17Claude Opus 4.8Opus 5 System Card capability summary · adaptive thinking · max effort unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.92.5%
18Claude Opus 4.7private validation · adaptive thinking · max effort · capability summary· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.92%
19GPT-5.2 ProVerified· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.90.5%
20GPT-5.2 ProVerified · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.90.5%
21Claude Sonnet 4.6private set · 120K thinking tokens · high effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.5%
22GPT 5.2Verified· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.2%
23GPT 5.2Verified · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.2%
24Claude Opus 4.5private validation · 64K thinking tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80%
25Step 3.5 FlashStep 3.5 technical report · Tables 12–13 · Python tool· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.56.5%
26Step 3.5 FlashStep 3.5 technical report · Tables 12–13 · vanilla· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.54.8%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.