CritPt v2, hard subset (Artificial Analysis run) leaderboard
- Claude Fable 5.2 leads
- 76.5%
- models measured
- 56
- lead over #2
- +4.5 pts
- median
- 33.8%
Leader on CritPt v2, hard subset (Artificial Analysis run) · read November 14
On November 14, 2026, 56 of our catalog's models had a CritPt v2, hard subset (Artificial Analysis run) result, and Anthropic Claude Fable 5.2 led with 76.5%, ahead of OpenAI GPT-6 Astra at 72.0%.
The hard subset of CritPt, research-level physics problems. Share solved. Run by Artificial Analysis on its own harness.
Role in the index
CritPt v2, hard subset (Artificial Analysis run) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2Anthropic | 76.5% | 87.8 | $40.00 | |
| 2 | GPT-6 AstraOpenAI | 72.0% | 83.8 | $50.00 | |
| 3 | GPT-6.1 SolOpenAI | 69.5% | 79.8 | $10.00 | |
| 4 | Gemini 4 CarbonGoogle | 69.0% | 86.6 | $10.00 | |
| 5 | Claude Opus 5.5Anthropic | 68.0% | 87.3 | $20.00 | |
| 6 | GPT-6.1 TerraOpenAI | 68.0% | 76.4 | $5.00 | |
| 7 | GPT-6 SolOpenAI | 66.5% | 72.1 | $10.00 | |
| 8 | Gemini 4 HydrogenGoogle | 66.0% | 77.9 | $7.50 | |
| 9 | Claude Fable 5.1Anthropic | 62.0% | 82.8 | $50.00 | |
| 10 | GPT-5.6 SolOpenAI | 60.0% | 70.6 | $20.00 | |
| 11 | Claude Sonnet 5.5Anthropic | 59.0% | 84.3 | $10.00 | |
| 12 | Gemini 4 ArgonGoogle | 53.5% | 80.3 | $10.00 | |
| 13 | Muse WonderMeta | 51.0% | 71.8 | $20.00 | |
| 14 | GPT-5.6 TerraOpenAI | 50.5% | 63.5 | $12.00 | |
| 15 | Claude Opus 5Anthropic | 50.0% | 81.7 | $25.00 | |
| 16 | Kimi K3.1 ProMoonshot AI | 49.5% | 79.0 | $10.00 | |
| 17 | Muse Spark 1.3Meta | 48.0% | 56.1 | $4.25 | |
| 18 | Claude Haiku 5.5Anthropic | 47.5% | 68.4 | $0.50 | |
| 19 | Gemini 3.8 FlashGoogle | 46.5% | 60.3 | $3.75 | |
| 20 | Grok 4.7xAI | 43.5% | 64.8 | $6.00 | |
| 21 | GPT-6 LunaOpenAI | 42.5% | 54.9 | $0.50 | |
| 22 | Kimi K3Moonshot AI | 40.5% | 61.4 | $15.00 | |
| 23 | Grok 4.6xAI | 40.0% | 63.9 | $6.00 | |
| 24 | Muse EmberMeta | 39.0% | 64.9 | $0.50 | |
| 25 | Hy4Tencent | 39.0% | 65.2 | $1.2505 | |
| 26 | Ling 3.1 FlashinclusionAI | 38.5% | Not benchmarked | $0.135 | |
| 27 | T Inkling 2 SmallThinking Machines | 35.0% | 48.9 | $1.00 | |
| 28 | MiMo-V2.6-ProXiaomi | 34.0% | 61.8 | $0.87 | |
| 29 | GLM 5.3Z.ai | 33.5% | 68.5 | $4.40 | |
| 30 | MiniMax M3.1 FlashMiniMax | 32.5% | 55.7 | $0.60 | |
| 31 | Step 5 PreviewStepFun | 32.5% | 57.4 | $2.70 | |
| 32 | GPT-5.6 LunaOpenAI | 30.5% | 55.3 | $1.20 | |
| 33 | Muse Spark OpenMeta | 29.5% | 43.4 | $4.25 | |
| 34 | Qwen3.8 Max (0902)Qwen | 29.5% | 67.3 | $6.00 | |
| 35 | MiMo-V2.6-FlashXiaomi | 29.5% | 54.8 | $0.28 | |
| 36 | DeepSeek V4 Pro 0813DeepSeek | 28.5% | 56.1 | $1.98 | |
| 37 | MiniMax M3MiniMax | 28.0% | 33.6 | $1.20 | |
| 38 | DeepSeek V4.1 Flash 1017DeepSeek | 27.5% | 66.9 | $0.60 | |
| 39 | Mistral Large 4Mistral AI | 27.0% | 60.9 | $2.10 | |
| 40 | GLM 5.3 FlashZ.ai | 24.5% | 56.2 | $0.50 | |
| 41 | T Inkling SmallThinking Machines | 18.5% | 27.3 | $1.20 | |
| 42 | T InklingThinking Machines | 18.0% | 29.2 | $4.05 | |
| 43 | Solar Pro 4Upstage | 18.0% | Not benchmarked | $0.60 | |
| 44 | Qwen3.8 27BQwen | 16.5% | 69.7 | $2.55 | |
| 45 | Nemotron 3 UltraNVIDIA | 15.0% | 30.0 | $2.30 | |
| 46 | Nemotron 3.5 Super VLNVIDIA | 13.5% | 14.8 | $0.50 | |
| 47 | GPT-5OpenAI | 11.5% | 38.8 | $10.00 | |
| 48 | Gemma 4 31BGoogle | 4.0% | 21.9 | $0.40 | |
| 49 | gpt-oss-120bOpenAI | 4.0% | Not benchmarked | $0.45 | |
| 50 | Gemini 3.5 Flash LiteGoogle | 3.5% | 48.9 | $2.50 | |
| 51 | Gemini 3.1 Flash LiteGoogle | 3.0% | Not benchmarked | $1.50 | |
| 52 | Command ACohere | 2.5% | Not benchmarked | $10.00 | |
| 53 | Nemotron 3.5 LightningNVIDIA | 1.5% | 22.6 | $0.18 | |
| 54 | Mistral Medium 3.5Mistral AI | 0.5% | 25.4 | $7.50 | |
| 55 | Granite 4.2 8BIBM Granite | 0.0% | Not benchmarked | $0.20 | |
| 56 | Mistral Small 4Mistral AI | 0.0% | Not benchmarked | $0.60 |
Only models in our catalog are listed, and each appears once with its best published run. 214 rows were read from the source on November 14, 2026, and rows that match no catalog model are left out.