The best at HLE
Claude Fable 5.2
80.2% on HLE
Today's AI models, ranked by how well they reason on two tests from an independent evaluator. One with expert-level questions across many subjects and one with physics research problems.
Data from 13 November 2026.
The best at HLE
Claude Fable 5.2
80.2% on HLE
The best at CritPt
Claude Fable 5.2
76.5% on CritPt
The best open-weights model
Kimi K3.1 Pro
37.3% on HLE
The cheapest per task
GPT-6.1 Terra
$0.96 per task, among the top ten
58 models
| # | Model | ||||||
|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 Anthropic · max effort | 80.2% | 76.5% | 65 | $26.54 | $8 and $40 | 9 Nov 2026 |
| 2 | Gemini 4 Carbon Google · high effort | 69.1% | 69.0% | 60 | $5.48 | $2 and $10 $4 and $20 from 1 December | 10 Nov 2026 |
| 3 | Claude Opus 5.5 Anthropic · max effort | 67.9% | 68.0% | 59 | $10.04 | $4 and $20 | 22 Sept 2026 |
| 4 | GPT-6 Astra OpenAI · max effort | 63.8% | 70.5% | 57 | $6.24 | $10 and $50 | 3 Sept 2026 |
| 5 | GPT-6.1 Sol OpenAI · max effort | 58.8% | 69.5% | 55 | $1.27 | $2 and $10 | 29 Sept 2026 |
| 6 | Gemini 4 Hydrogen Google · extra high effort | 58.8% | 66.0% | 54 | $3.16 | $1.50 and $7.50 | 12 Nov 2026 |
| 7 | Gemini 4 Argon Google · high effort | 56.7% | 52.5% | 50 | $4.97 | $2 and $10 $4 and $20 from 1 December | 13 Oct 2026 |
| 8 | Claude Sonnet 5.5 Anthropic · max effort | 53.3% | 59.0% | 55 | $10.02 | $2 and $10 | 28 Sept 2026 |
| 9 | Muse Wonder Full Meta · max effort | 48.5% | 56.5% | 50 | No data | No per-token price | 12 Oct 2026 |
| 10 | GPT-6.1 Terra OpenAI · max effort | 47.3% | 68.0% | 51 | $0.96 | $1 and $5 | 26 Oct 2026 |
| 11 | Muse Wonder Meta · max effort | 39.0% | 50.5% | 44 | $6.95 | $4 and $20 | 12 Oct 2026 |
| 12 | Kimi K3.1 Pro Moonshot AI · open weights | 37.3% | 49.5% | 46 | $3.68 | $2 and $10 | 24 Oct 2026 |
| 13 | Gemini 3.8 Flash Google · high effort | 33.1% | 46.5% | 31 | $7.77 | $0.75 and $3.75 $1.50 and $7.50 from 1 January 2027 | 2 Sept 2026 |
| 14 | Tencent Hy4 Tencent | 32.7% | 39.0% | 35 | $1.03 | $0.834 and $2.501 half price until 31 December | 4 Nov 2026 |
| 15 | Claude Haiku 5.5 Anthropic · max effort | 29.6% | 43.5% | 36 | $0.41 | $0.10 and $0.50 | 7 Oct 2026 |
| 16 | Inkling 2 Small Thinking Machines · open weights | 28.7% | 35.0% | 31 | $1.18 | $0.30 and $1 | 5 Nov 2026 |
| 17 | MiniMax-M3.1-Flash MiniMax · open weights | 28.1% | 32.5% | 29 | $0.49 | $0.15 and $0.60 | 19 Oct 2026 |
| 18 | Muse Spark Open Meta · extra high effort · open weights | 26.3% | 29.5% | 25 | $2.02 | No per-token price | 15 Oct 2026 |
| 19 | DeepSeek V4.1 Flash 1017 DeepSeek · max effort · open weights | 25.3% | 25.0% | 33 | $0.60 | $0.30 and $1.20 | 17 Oct 2026 |
| 20 | Muse Ember Meta · max effort | 25.1% | 39.0% | 36 | $0.14 | $0.10 and $0.50 | 9 Nov 2026 |
| 21 | Step 5 Preview StepFun · open weights | 24.3% | 32.5% | 29 | $2.35 | $1 and $2.70 | 20 Sept 2026 |
| 22 | Qwen3.8 Max (0902) Alibaba | 24.0% | 29.5% | 34 | $10.02 | $2 and $6 | 2 Sept 2026 |
| 23 | Grok 4.7 xAI · extra high effort | 23.2% | 33.5% | 35 | $7.77 | $2 and $6 | 21 Sept 2026 |
| 24 | GPT-6 Luna OpenAI · max effort | 22.2% | 42.5% | 28 | $0.15 | $0.10 and $0.50 | 22 Sept 2026 |
| 25 | DeepSeek V4 Pro 0813 DeepSeek · max effort · open weights | 19.5% | 23.5% | 23 | $1.77 | $1.32 and $3.96 | 13 Aug 2026 |
| 26 | GLM-5.3-Flash Z.ai · open weights | 18.2% | 24.5% | 28 | $0.58 | $0.15 and $0.50 | 26 Aug 2026 |
| 27 | GLM-5.3 Z.ai · max effort · open weights | 17.9% | 33.5% | 32 | $5.01 | $1.40 and $4.40 | 18 Aug 2026 |
| 28 | Mistral Large 4 Mistral · high effort · open weights | 17.3% | 27.0% | 26 | $2.06 | $0.70 and $2.10 | 6 Oct 2026 |
| 29 | Ling 3.1 Flash 1019 InclusionAI · open weights | 17.2% | 38.5% | 30 | $3.19 | No per-token price | No data |
| 30 | Agnes 2.5 Pro Theta Sapiens AI | 16.8% | 26.0% | 24 | $0.57 | $0.10 and $0.30 | No data |
| 31 | JT-4.1 Flash 236B A21B China Mobile | 16.4% | 19.0% | 19 | No data | No per-token price | No data |
| 32 | Qwen3.8-Flash-Next Alibaba · open weights | 16.3% | 24.5% | 29 | $0.57 | No per-token price | 26 Aug 2026 |
| 33 | MiniMax-M3 MiniMax · open weights | 16.0% | 28.0% | 19 | $1.29 | $0.30 and $1.20 | 1 Jun 2026 |
| 34 | Qwen3.8 27B Alibaba · extra high effort · open weights | 15.3% | 4.0% | 20 | $3.53 | $0.50 and $3 | 14 Aug 2026 |
| 35 | MiMo-V2.6-Pro Xiaomi · open weights | 15.2% | 34.0% | 30 | $0.26 | $0.435 and $0.87 | 22 Sept 2026 |
| 36 | MiMo-V2.6-Flash Xiaomi · open weights | 13.7% | 29.5% | 27 | $0.13 | $0.14 and $0.28 | 22 Sept 2026 |
| 37 | Beam 1.0 Reflection · open weights | 12.8% | 26.5% | 18 | $0.28 | No per-token price | 16 Oct 2026 |
| 38 | Inkling Small Thinking Machines · open weights | 12.1% | 18.5% | 13 | $0.20 | $0.30 and $1.20 | 30 Jul 2026 |
| 39 | Apodex 1.1 Apodex | 11.6% | 17.0% | 14 | $0.76 | $0.60 and $3.60 | 24 Aug 2026 |
| 40 | Inkling Thinking Machines · extra high effort · open weights | 10.6% | 18.0% | 12 | $1.35 | $1 and $4.05 | 15 Jul 2026 |
| 41 | K2 Horizon 375B A23B MBZUAI IFM · open weights | 10.2% | 12.0% | 18 | $0.00 | No per-token price | No data |
| 42 | Gemini 3.5 Flash-Lite Google | 9.9% | 3.5% | 12 | $0.69 | $0.30 and $2.50 | 21 Jul 2026 |
| 43 | Muse Glimmer Meta · high effort · open weights | 9.1% | 7.0% | 6 | $0.17 | No per-token price | 10 Aug 2026 |
| 44 | Nemotron 3.5 Super VL NVIDIA · open weights | 8.8% | 13.5% | 14 | $0.52 | No per-token price | 15 Oct 2026 |
| 45 | Nemotron 3 Ultra 550B A55B NVIDIA · open weights | 8.5% | 15.0% | 12 | $2.23 | No per-token price | 4 Jun 2026 |
| 46 | Gemma 4 31B Google · open weights | 7.2% | 4.0% | 4 | No data | No per-token price | 2 Apr 2026 |
| 47 | Nanbeige4.3-33B Nanbeige · open weights | 6.9% | 15.5% | 20 | No data | No per-token price | No data |
| 48 | K-EXAONE 2.0 LG AI Research · open weights | 5.7% | 4.0% | 8 | No data | No per-token price | 31 Jul 2026 |
| 49 | Command A+ Cohere · open weights | 5.1% | 2.5% | 6 | $0.00 | Free | 20 May 2026 |
| 50 | Solar Pro 4 Upstage | 4.7% | 18.0% | 14 | $0.96 | $0.30 and $1.20 | 6 Aug 2026 |
| 51 | Mistral Medium 3.5 Mistral · open weights | 4.7% | 0.5% | 5 | $1.04 | $1.50 and $7.50 | 28 Apr 2026 |
| 52 | Nemotron 3.5 Lightning NVIDIA · open weights | 4.5% | 1.5% | 6 | $0.31 | No per-token price | 11 Aug 2026 |
| 53 | Mistral Small 4 Mistral · open weights | 4.4% | 0.0% | 4 | $0.04 | $0.15 and $0.60 | 16 Mar 2026 |
| 54 | North Mini Code Cohere · open weights | 4.2% | 1.0% | 3 | $0.00 | Free | 9 Jun 2026 |
| 55 | Granite 4.2 8B IBM · open weights | 4.0% | 0.0% | 5 | $0.04 | No per-token price | 25 Aug 2026 |
| 56 | gpt-oss-120b OpenAI · high effort · open weights | 3.8% | 4.0% | 4 | $0.35 | No per-token price | 5 Aug 2025 |
| 57 | Granite 4.2 3B IBM · open weights | 2.9% | 0.0% | 5 | No data | No per-token price | 25 Aug 2026 |
| 58 | NVIDIA Nemotron 3 Nano 4B NVIDIA · open weights | 2.8% | 0.0% | 2 | No data | No per-token price | 16 Mar 2026 |
HLE Diamond and CritPt v2's hard subset, the score and the cost per task from Artificial Analysis, read on 13 November 2026 with each model at its highest effort. The price and the release date from each maker's page. HLE Diamond and CritPt v2 on Artificial Analysis.
AI · Nov 13, 2026
Case study · Nov 12, 2026
Case study · Nov 12, 2026
Case study · Nov 11, 2026
Case study · Nov 11, 2026
AI · Oct 26, 2026
Case study · Nov 10, 2026
Case study · Nov 10, 2026
Case study · Nov 11, 2026
Software · Oct 25, 2026
Case study · Nov 10, 2026No single one wins at everything. We explain it in the article “Which AI Model to Use to Build Your Company's Software”, including how we split the work between them.
By Artificial Analysis's tests on 13 November 2026, Claude Fable 5.2, from Anthropic, which wins at both, with 80.2% of HLE Diamond and 76.5% of CritPt's hard subset.
Whether the model can answer 1,000 expert-level questions in mathematics, science and the humanities, half of them about reasoning and half about knowledge. It is a revised version of Humanity's Last Exam, from the Center for AI Safety and Scale AI.
Whether the model can solve physics research problems with a computer to run code and no internet. It comes from Argonne and the University of Illinois, and we show its hardest part, which Artificial Analysis publishes on its own.
What it costs on average for the model to do one of the tasks behind Artificial Analysis's score, across all its tests. It says more than the price per million tokens, because for the same work some models use many more tokens than others.
In “AI comparison 2026”, with every model ranked by its score, what each task costs and how long it takes, its price and its context.