All tests
Reasoning
LiveBench reasoning
Logical, spatial and multi-step reasoning tasks.
Published by LiveBench under Apache-2.0. Compare it with other tests
Top score92.7%GPT-6 Astra (max), OpenAI
Models tested63from 11 makers
Median score85.8%Half the models score above it
Scale0 to 100%The share of tasks done right
Top 15
Each model at its best configuration, coloured by maker.
The best model from each maker
Each maker’s highest-scoring model on this test, and its rank.
| Maker | Best model | Score | Rank |
|---|---|---|---|
 OpenAI | GPT-6 Astra (max) | 92.7% | #1 of 63 |
 Anthropic | Claude Opus 5.5 (max effort) | 92.2% | #3 of 63 |
 Moonshot AI | Kimi K3 | 90.7% | #8 of 63 |
 xAI | Grok 4.6 | 90.5% | #10 of 63 |
 Google | Gemini 3.8 Flash (high) | 89.3% | #16 of 63 |
 Alibaba | Qwen3.8 Max | 88.2% | #21 of 63 |
 DeepSeek | DeepSeek V4.1 Flash (max) | 86.7% | #28 of 63 |
 Z.ai | GLM-5.3 | 85.8% | #32 of 63 |
 NVIDIA | Nemotron 3 Ultra 550B A55B | 74.7% | #57 of 63 |
 MiniMax | MiniMax-M3 | 74.5% | #58 of 63 |
Every result
63 models, each once at its best configuration, highest score first. Models with the same score share a rank.
| Rank | Model | Maker | Score |
|---|---|---|---|
| #1 | GPT-6 Astra (max) | OpenAI | 92.7% |
| #2 | GPT-6.1 Sol (max) | OpenAI | 92.6% |
| #3 | Claude Opus 5.5 (max effort) | Anthropic | 92.2% |
| #4 | Claude Fable 5.1 (max effort) | Anthropic | 91.7% |
| #5 | GPT-5.6 Sol (max) | OpenAI | 91.7% |
| #6 | Claude Sonnet 5.5 (max effort) | Anthropic | 91.6% |
| #7 | Claude Opus 5 (max effort) | Anthropic | 91.2% |
| #8 | Kimi K3 | Moonshot AI | 90.7% |
| #9 | GPT-5.6 Terra (max) | OpenAI | 90.6% |
| #10 | Grok 4.6 | xAI | 90.5% |
| #11 | Smaug Agentic |  Not reported | 90.3% |
| #12 | Muse Spark 1.2 (xhigh) | Meta AI | 90.0% |
| #13 | Claude Fable 5 (max effort) | Anthropic | 89.7% |
| #13 | GPT-5.5 (xhigh) | OpenAI | 89.7% |
| #13 | Muse Spark 1.3 (xhigh) | Meta AI | 89.7% |
| #16 | Gemini 3.8 Flash (high) | Google | 89.3% |
| #17 | Claude Opus 4.8 (max effort) | Anthropic | 89.2% |
| #18 | Claude Sonnet 5 (xhigh effort) | Anthropic | 88.7% |
| #19 | Claude Opus 4.6 (thinking auto high effort) | Anthropic | 88.7% |
| #20 | GPT-6 Sol (max) | OpenAI | 88.7% |
| #21 | Qwen3.8 Max | Alibaba | 88.2% |
| #22 | GPT-5.4 (xhigh) | OpenAI | 88.1% |
| #23 | Gemini 3.7 Flash (high) | Google | 87.8% |
| #24 | Muse Spark 1.1 (xhigh) | Meta AI | 87.7% |
| #25 | Qwen3.8 Flash Next | Alibaba | 87.4% |
How it is scored
Arithmetic mean of the 4 published task scores in this category. Only complete results are ranked. Exact model and effort configurations are preserved.
Results from LiveBench (Apache-2.0), as published. Model names link to their pages where the site can match them; makers come from the model’s page, or else from the publisher.