New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
LiveBench benchmark
AI model leaderboard for the LiveBench benchmark. Compare how large language models score on LiveBench, see the full ranking, and understand what this AI benchmark measures. Qwen3-235B-A22B currently leads with 77.1. Contamination-resistant general benchmark with monthly question rotation across reasoning, coding, math, language.
Leaderboard
| # | Model | Organization | Score | Variant | Source |
|---|---|---|---|---|---|
| #1 | Qwen3-235B-A22B | Qwen | 77.1 | thinking-2024-11-25 | official ↗ |
| #2 | Qwen3-32B | Qwen | 74.9 | thinking-2024-11-25 | official ↗ |
| #3 | Qwen3-30B-A3B | Qwen | 74.3 | thinking-2024-11-25 | official ↗ |
| #4 | Qwen3-14B | Qwen | 71.3 | thinking-2024-11-25 | official ↗ |
| #5 | Qwen3-8B | Qwen | 67.1 | thinking-2024-11-25 | official ↗ |
| #6 | Qwen3-4B | Qwen | 63.6 | thinking-2024-11-25 | official ↗ |
| #7 | Qwen3-235B-A22B | Qwen | 62.5 | non-thinking-2024-11-25 | official ↗ |
| #8 | Qwen3-32B | Qwen | 59.8 | non-thinking-2024-11-25 | official ↗ |
| #9 | Qwen3-14B | Qwen | 59.6 | non-thinking-2024-11-25 | official ↗ |
| #10 | Qwen3-30B-A3B | Qwen | 59.4 | non-thinking-2024-11-25 | official ↗ |
| #11 | Qwen3-8B | Qwen | 53.5 | non-thinking-2024-11-25 | official ↗ |
| #12 | Qwen3-1.7B | Qwen | 51.1 | thinking-2024-11-25 | official ↗ |
| #13 | Qwen2.5-32B | Qwen | 50.6 | instruct-cited-exaone35 | official ↗ |
| #14 | Qwen3-4B | Qwen | 48.4 | non-thinking-2024-11-25 | official ↗ |
| #15 | Kimi Linear | Moonshot | 45.2 | sft-pass@1 | official ↗ |
| #16 | EXAONE 3.5 32B | LG AI Research | 43 | instruct-from-hf-readme | official ↗ |
| #17 | Gemma 2 27B | Google DeepMind | 40 | instruct-cited-exaone35 | official ↗ |
| #18 | EXAONE 3.5 7.8B | LG AI Research | 39.8 | instruct-from-hf-readme | official ↗ |
| #19 | Qwen2.5-7B | Qwen | 35.6 | instruct-cited-exaone35-7b | official ↗ |
| #20 | Qwen3-1.7B | Qwen | 35.6 | non-thinking-2024-11-25 | official ↗ |
| #21 | EXAONE 3.5 2.4B | LG AI Research | 33 | instruct-from-hf-readme | official ↗ |
| #22 | Gemma 2 9B | Google DeepMind | 32.1 | instruct-cited-exaone35 | official ↗ |
| #23 | Qwen3-0.6B | Qwen | 30.3 | thinking-2024-11-25 | official ↗ |
| #24 | Llama 3.1-8B | Meta AI | 28.3 | instruct-cited-exaone35 | official ↗ |
| #25 | Yi-1.5-34B | 01.AI | 26.2 | instruct-cited-exaone35 | official ↗ |
| #26 | Qwen2.5-3B | Qwen | 25.7 | instruct-cited-exaone35 | official ↗ |
| #27 | Llama 3.2 3B | Meta AI | 24 | instruct-cited-exaone35 | official ↗ |
| #28 | Qwen3-0.6B | Qwen | 21.8 | non-thinking-2024-11-25 | official ↗ |
| #29 | Gemma 2 2B | Google DeepMind | 20 | instruct-cited-exaone35 | official ↗ |
| #30 | Qwen2.5-1.5B | Qwen | 19.2 | instruct-cited-exaone35 | official ↗ |
Frequently asked questions about LiveBench
What is the LiveBench benchmark?
Contamination-resistant general benchmark with monthly question rotation across reasoning, coding, math, language.
How is the LiveBench benchmark scored?
LiveBench is scored using the accuracy metric, where a higher score is better. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.
Which AI model scores highest on LiveBench?
As of the latest reported scores on GenAIList, Qwen3-235B-A22B achieves the highest result on LiveBench with a score of 77.1.
Is a higher LiveBench score better?
Yes. On LiveBench a higher score indicates better performance, so models near the top of the leaderboard are the strongest.