BlogLLM Benchmark Scores as Published, Side by Side
ARC-AGI-2: Results as Published
Every result the catalog holds for ARC-AGI-2: 20 models from 3 developers, read from 10 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area General Knowledge and ReasoningMaintainer ARC Prize FoundationIntroduced 2025Data version 2026103001, as of 2026-10-30
ARC-AGI-2 is a harder successor to ARC-AGI-1 that keeps the same kind of puzzle but is built so that brute-force compute does not solve it. It measures how well a system learns new rules from a few examples. At launch in March 2025, human testers averaged about 60 percent, while leading AI reasoning models scored about 1 percent.
ARC-AGI-2
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6 AstraOpenAI, released 2026-09-03 | 95% | OpenAI, GPT-6 Astrachecked 2026-10-14 | - |
| GPT-5.6 SolOpenAI, released 2026-07-09 | 92.5% | OpenAI, GPT-6 Astrachecked 2026-10-14; printed on 1 more page | - |
| GPT-5.5OpenAI, released 2026-04-23 | 85% | OpenAI, Introducing GPT-5.5checked 2026-10-15; printed on 1 more page | Effort: xhigh |
| GPT-5.4OpenAI, released 2026-03-05 | 73.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh |
| GPT-5.4 ProOpenAI, released 2026-03-05 | 83.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh |
| GPT-5.2OpenAI, released 2025-12-11 | 52.9% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 2 more pages | Effort: xhigh |
| GPT-5.2 ProOpenAI, released 2025-12-11 | 54.2% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 3 more pages | Effort: high |
| GPT-5.1OpenAI, released 2025-11-12 | 17.6% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 1 more page | Effort: high |
| Claude Fable 5.1Anthropic, released 2026-09-01 | 90% | Anthropic, System Card: Claude Fable 5.1 & Claude Mythos 5.1checked 2026-10-25; printed on 1 more page | Effort: max |
| Claude Opus 5Anthropic, released 2026-07-24 | 90.42% | Anthropic, System Card: Claude Fable 5.1 & Claude Mythos 5.1checked 2026-10-25; printed on 1 more page | Effort: max |
| Claude Fable 5Anthropic, released 2026-06-09 | 89.2% | Anthropic, System Card: Claude Fable 5.1 & Claude Mythos 5.1checked 2026-10-25; printed on 1 more page | Effort: max |
| Claude Opus 4.7Anthropic, released 2026-04-16 | 75.8% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15; printed on 1 more page | - |
| Claude Sonnet 4.6Anthropic, released 2026-02-17 | 60.4% | Anthropic, Claude Sonnet 4.6checked 2026-10-15; printed on 2 more pages | Effort: high |
| 58.3% | Anthropic, Claude Sonnet 4.6checked 2026-10-15 | Effort: max | |
| Claude Opus 4.6Anthropic, released 2026-02-05 | 68.8% | Anthropic, Claude Opus 4.6checked 2026-10-15; printed on 2 more pages | Effort: max |
| Claude Opus 4.5Anthropic, released 2025-11-24 | 37.6% | Anthropic, Claude Opus 4.5checked 2026-10-15; printed on 2 more pages | Effort: high |
| Claude Sonnet 4.5Anthropic, released 2025-09-29 | 13.6% | Anthropic, Claude Opus 4.5checked 2026-10-15; printed on 2 more pages | - |
| Gemini 3.5 FlashGoogle, released 2026-05-19 | 72.1% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | - |
| Gemini 3.1 Pro PreviewGoogle, released 2026-02-19 | 77.1% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 2 more pages | Effort: high |
| Gemini 3 FlashGoogle, released 2025-12-17 | 33.6% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | - |
| Gemini 3 ProGoogle, released 2025-11-18 | 31.1% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 3 more pages | Effort: high |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.