Best LLM for reasoning: Reasoning Index ranking
- Gemini 4 Argon leads
- 90.7
- models measured
- 122
- lead over #2
- +0.1
- median
- 60.3
Leader on Reasoning Index · read October 22
As of October 22, 2026, Google Gemini 4 Argon leads reasoning with 90.7 on Reasoning Index, ahead of Anthropic Claude Opus 5.5 (90.6) and OpenAI GPT-5.4 Pro (90.1).
| Rank | Model | Reasoning Index | Evidence | GPQA Diamond | Humanity's Last Exam | LiveBench Reasoning | ARC-AGI-2 | Output $/Mtok |
|---|---|---|---|---|---|---|---|---|
| 1 | Gemini 4 Argon Google | 90.7 | 96.4% | 57.1% | 95.1% | 91.3% | $10.00 | |
| 2 | Claude Opus 5.5 Anthropic | 90.6 | 90.6% | 61.4% | 92.2% | 93.3% | $20.00 | |
| 3 | GPT-5.4 Pro OpenAI | 90.1 | 94.6% | No result | No result | 83.3% | $180.00 | |
| 4 | GPT-6 Astra OpenAI | 89.7 | 95.8% | 54.7% | 92.7% | 95.0% | $50.00 | |
| 5 | GPT-6.1 Sol OpenAI | 88.6 | 95.4% | 52.9% | 92.6% | 94.2% | $10.00 | |
| 6 | Claude Sonnet 5.5 Anthropic | 88.2 | 95.6% | 55.0% | 91.6% | No result | $10.00 | |
| 7 | Claude Fable 5.1 Anthropic | 88.1 | No result | 59.1% | 91.7% | 90.0% | $50.00 | |
| 8 | Claude Opus 5 Anthropic | 88.0 | 93.9% | 54.9% | 91.2% | 90.4% | $25.00 | |
| 9 | GPT-5.6 Sol OpenAI | 86.1 | 93.5% | 49.5% | 91.7% | 92.5% | $20.00 | |
| 10 | Claude Fable 5 Anthropic | 85.3 | 85.9% | 55.5% | 89.7% | 89.2% | $50.00 | |
| 11 | Gemini 3.8 Flash Google | 84.8 | 95.4% | 47.8% | 89.3% | 89.2% | $3.75 | |
| 12 | GPT-6 Sol OpenAI | 84.5 | 94.3% | 47.9% | 88.7% | 89.6% | $10.00 | |
| 13 | Gemini 3.7 Flash Google | 83.5 | 94.8% | 47.9% | 87.8% | 84.6% | $3.75 | |
| 14 | GPT-5.5 OpenAI | 81.8 | 90.7% | 45.8% | 89.7% | 85.0% | $30.00 | |
| 15 | GPT-5.6 Terra OpenAI | 81.3 | 93.3% | 42.9% | 90.6% | 83.9% | $12.00 | |
| 16 | Claude Opus 4.8 Anthropic | 80.4 | 91.0% | 48.7% | 89.2% | 72.1% | $25.00 | |
| 17 | DeepSeek V4.1 Flash 1017 DeepSeek | 80.4 | 93.2% | 43.3% | 89.1% | No result | $0.60 | |
| 18 | Gemini 3.1 Pro Preview Google | 79.9 | 94.4% | 47.0% | No result | 77.1% | $12.00 | |
| 19 | Qwen3.8 Max (0902) Qwen | 79.9 | 92.7% | 43.1% | 88.2% | No result | $6.00 | |
| 20 | Grok 4.6 xAI | 78.8 | 94.0% | 44.1% | 91.2% | 67.1% | $6.00 | |
| 21 | Kimi K3 Moonshot AI | 78.3 | 93.1% | 46.9% | 90.7% | 60.4% | $15.00 | |
| 22 | Claude Sonnet 5 Anthropic | 78.2 | 90.5% | 41.3% | 88.7% | No result | $10.00 | |
| 23 | GLM 5.3 Z.ai | 78.2 | 90.9% | 42.3% | 85.8% | No result | $4.40 | |
| 24 | Muse Spark 1.3 Meta | 77.6 | No result | 48.7% | 89.7% | No result | $4.25 | |
| 25 | GPT-5.4 OpenAI | 76.9 | 93.3% | 43.7% | No result | 74.0% | $15.00 | |
| 26 | DeepSeek V4.1 Flash DeepSeek | 75.7 | 89.8% | 39.2% | 86.7% | 72.9% | $0.84 | |
| 27 | Gemini 3.5 Flash Google | 75.7 | 92.8% | 42.7% | No result | 72.1% | $9.00 | |
| 28 | Claude Opus 4.7 Anthropic | 75.4 | 90.2% | 42.3% | No result | 75.8% | $25.00 | |
| 29 | Grok 4.7 xAI | 75.1 | 92.7% | 43.1% | 82.7% | 61.4% | $6.00 | |
| 30 | Muse Spark 1.2 Meta | 75.0 | No result | 45.5% | 90.0% | No result | $4.25 | |
| 31 | Claude Haiku 5.5 Anthropic | 74.7 | 89.6% | 44.4% | 82.4% | 61.3% | $0.50 | |
| 32 | Qwen3.7 Max Qwen | 74.4 | 90.9% | 40.5% | No result | No result | $4.425 | |
| 33 | Grok 4.5 xAI | 74.3 | 93.4% | 42.7% | 87.2% | 52.6% | $6.00 | |
| 34 | MiMo-V2.6-Pro Xiaomi | 74.2 | No result | 49.4% | 79.6% | No result | $0.87 | |
| 35 | MiniMax M3 MiniMax | 73.6 | 90.9% | 39.0% | 74.5% | No result | $1.20 | |
| 36 | GPT-5.6 Luna OpenAI | 73.4 | 91.6% | 39.5% | 85.6% | 59.5% | $1.20 | |
| 37 | Step 5 Preview StepFun | 73.4 | No result | 46.5% | 83.8% | No result | $2.70 | |
| 38 | GLM 5.3 Flash Z.ai | 72.8 | 90.2% | 39.9% | 77.6% | 65.8% | $0.50 | |
| 39 | Mistral Large 4 Mistral AI | 72.8 | 88.1% | 35.0% | 83.9% | No result | $4.18 | |
| 40 | Claude Opus 4.6 Anthropic | 72.6 | 90.5% | 39.9% | No result | 69.2% | $25.00 | |
| 41 | Gemini 3.6 Flash Google | 72.3 | 94.1% | 40.8% | No result | 60.4% | $3.75 | |
| 42 | Kimi K2.6 Moonshot AI | 72.2 | 90.8% | 37.5% | No result | No result | $4.00 | |
| 43 | GPT-6 Luna OpenAI | 71.9 | 90.5% | 38.5% | 81.8% | 59.3% | $0.50 | |
| 44 | DeepSeek V4 Pro 0813 DeepSeek | 71.7 | 91.7% | 41.0% | No result | 61.3% | $1.98 | |
| 45 | Grok 4.3 xAI | 71.0 | 88.8% | 37.2% | No result | No result | $2.50 | |
| 46 | DeepSeek V4 Flash 0731 DeepSeek | 70.2 | 91.0% | 38.6% | No result | 61.4% | $0.396 | |
| 47 | Claude Sonnet 4.6 Anthropic | 70.0 | 87.4% | 33.6% | 86.4% | 60.4% | $15.00 | |
| 48 | Qwen3.7 Plus Qwen | 69.4 | 87.9% | 35.6% | No result | No result | $1.28 | |
| 49 | Kimi K2.7 Code Moonshot AI | 68.9 | 87.9% | 35.0% | No result | No result | $4.00 | |
| 50 | GPT-5.2 OpenAI | 67.7 | 91.4% | 37.7% | No result | 52.9% | $14.00 | |
| 51 | Muse Spark Open Meta | 67.1 | No result | 43.9% | 73.7% | No result | $4.25 | |
| 52 | GLM 5.1 Z.ai | 66.4 | 89.9% | 30.1% | No result | No result | $4.40 | |
| 53 | Qwen3.6 Plus Qwen | 64.0 | 88.4% | 27.8% | No result | No result | $1.95 | |
| 54 | Qwen3.5 397B A17B Qwen | 63.9 | 86.4% | 29.0% | No result | No result | $2.34 | |
| 55 | DeepSeek V4 Flash Vision Exp DeepSeek | 63.7 | No result | 34.5% | 85.4% | No result | $1.32 | |
| 56 | T Inkling Thinking Machines | 63.1 | 88.3% | 31.9% | 78.3% | 36.5% | $4.05 | |
| 57 | Nemotron 3 Ultra NVIDIA | 63.0 | 85.4% | 28.4% | No result | No result | $2.40 | |
| 58 | GLM 5.2 Z.ai | 62.2 | 91.9% | 41.1% | No result | 22.8% | $4.40 | |
| 59 | Gemini 3 Flash Preview Google | 61.5 | 89.4% | 36.6% | No result | 33.6% | $3.00 | |
| 60 | GLM 4.7 Z.ai | 61.2 | 83.3% | 27.4% | No result | No result | $2.20 | |
| 61 | T Inkling Small Thinking Machines | 61.1 | 88.5% | 33.3% | No result | 40.1% | $1.20 | |
| 62 | Qwen3.6 27B Qwen | 59.4 | 85.9% | 23.1% | No result | No result | $2.70 | |
| 63 | Qwen3.6 35B A3B Qwen | 58.3 | 84.8% | 22.2% | No result | No result | $1.00 | |
| 64 | Gemma 4 31B Google | 58.1 | 75.8% | 23.6% | 68.3% | No result | $0.40 | |
| 65 | Claude Opus 4.5 Anthropic | 57.8 | 86.0% | 30.1% | No result | 37.6% | $25.00 | |
| 66 | Nemotron 3.5 Super VL NVIDIA | 57.8 | No result | 30.7% | 78.8% | No result | $0.50 | |
| 67 | Qwen3.5-35B-A3B Qwen | 56.7 | 83.5% | 21.0% | No result | No result | $1.30 | |
| 68 | Gemini 2.5 Pro Preview 06-05 Google | 56.6 | 84.8% | 19.9% | No result | No result | $10.00 | |
| 69 | Qwen3.8 27B Qwen | 55.7 | No result | 33.9% | 80.0% | 42.4% | $2.55 | |
| 70 | Grok 4.20 xAI | 55.6 | No result | 34.5% | No result | 65.1% | $2.50 | |
| 71 | Claude Sonnet 4.5 Anthropic | 54.8 | 82.3% | No result | No result | 13.6% | $15.00 | |
| 72 | Gemini 3.1 Flash Lite Google | 53.2 | 81.8% | 17.2% | No result | No result | $1.50 | |
| 73 | GPT-5.1 OpenAI | 52.5 | 87.6% | 28.5% | No result | 17.6% | $10.00 | |
| 74 | GPT-5.4 Mini OpenAI | 52.4 | 86.9% | 28.1% | No result | 18.9% | $4.50 | |
| 75 | gpt-oss-120b OpenAI | 51.9 | 75.8% | 19.6% | No result | No result | $0.50 |
Showing the first 75 of 122. Every benchmark result, with its source date, is in the full dataset.
What this measures
Can the model work through hard expert questions, abstract puzzles and logic that searching the web does not answer?
The Reasoning Index averages four live benchmarks with the weights below. Each result is rescaled to 0 to 100 from a fixed range; a model missing a benchmark is scored on the ones it has, with the weights renormalised.
| Benchmark | Weight | Rescaled from | Source |
|---|---|---|---|
| GPQA Diamond | 30% | 0% to 100% | epoch.ai |
| Humanity's Last Exam | 30% | 0% to 70% | artificialanalysis.ai |
| LiveBench Reasoning | 20% | 0% to 100% | livebench.ai |
| ARC-AGI-2 | 20% | 0% to 100% | epoch.ai |
A model needs at least two of these benchmarks to be ranked. One result is a single reading, so a model with only one is listed under "Not enough data yet" below, with the benchmark it has. Each model uses its best published setting per benchmark. Read October 22, 2026. How the index is built
Not enough data yet
A score needs at least two of the Reasoning Index benchmarks. These 91 models have one, so they are not ranked.