Skip to content
Weights Not Walls
Menu

LiveBench

Reported

Questions in mathematics, coding, reasoning, data analysis, language and instruction following, refreshed regularly so that they are less likely to be in training data.

Kind
Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
Link
https://livebench.ai/#/
Score shown
Epoch's column "Global average", as a percentage, higher is better. The source gives no error bars.
Results
54 in all: 22 for open-weight models, 28 for closed ones, 4 where Epoch does not say. 19 linked to a model page here.

Best open-weight result: Qwen/QwQ-32B, 72.0%, number 6 of 54 results.

LiveBench results
#ModelWeightsSettingScoreDateSource
1gemini-2.5-pro-exp-03-25Google DeepMindcloseddefault82.3%not datedLiveBench Leaderboard
2gpt-5.1-2025-11-13_highOpenAIclosedeffort high78.8%not datedLiveBench Leaderboard
3claude-3-7-sonnet-20250219Anthropiccloseddefault76.1%not datedLiveBench Leaderboard
4o3-mini-2025-01-31_highOpenAIclosedeffort high75.9%not datedLiveBench Leaderboard
5o1-2024-12-17_highOpenAIclosedeffort high75.7%not datedLiveBench Leaderboard
6Qwen/QwQ-32BEpoch: QwQ-32B, Alibabaopendefault72.0%not datedLiveBench Leaderboard
7deepseek-ai/DeepSeek-R1Epoch: DeepSeek-R1, DeepSeekopendefault71.6%not datedLiveBench Leaderboard
8o3-mini-2025-01-31_mediumOpenAIclosedeffort medium70.0%not datedLiveBench Leaderboard
9gpt-4.5-preview-2025-02-27OpenAIcloseddefault69.0%not datedLiveBench Leaderboard
10gemini-2.0-flash-thinking-exp-01-21Google DeepMind,Googlecloseddefault66.9%not datedLiveBench Leaderboard
11deepseek-ai/DeepSeek-V3-0324Epoch: DeepSeek-V3-0324, DeepSeekopendefault66.9%not datedLiveBench Leaderboard
12claude-3-7-sonnet-20250219Anthropiccloseddefault65.6%not datedLiveBench Leaderboard
13gemini-2.0-pro-exp-02-05Google DeepMindcloseddefault65.1%not datedLiveBench Leaderboard
14gemini-exp-1206Google DeepMind,Googlecloseddefault64.1%not datedLiveBench Leaderboard
15o3-mini-2025-01-31_lowOpenAIclosedeffort low62.5%not datedLiveBench Leaderboard
16qwen2.5-maxAlibabacloseddefault62.3%not datedLiveBench Leaderboard
17gemini-2.0-flash-001Google DeepMind,Googlecloseddefault61.5%not datedLiveBench Leaderboard
18deepseek-ai/DeepSeek-V3Epoch: DeepSeek-V3, DeepSeekopendefault60.5%not datedLiveBench Leaderboard
19gemini-2.0-flash-expGoogle DeepMind,Googlecloseddefault59.3%not datedLiveBench Leaderboard
20claude-3-5-sonnet-20241022Anthropiccloseddefault59.0%not datedLiveBench Leaderboard
21o1-mini-2024-09-12_mediumOpenAIclosedeffort medium57.8%not datedLiveBench Leaderboard
22gpt-4o-2024-08-06OpenAIcloseddefault55.3%not datedLiveBench Leaderboard
23deepseek-ai/DeepSeek-R1-Distill-Llama-70BEpoch: DeepSeek-R1-Distill-Llama-70B, DeepSeekopendefault54.5%not datedLiveBench Leaderboard
24grok-2-1212xAIcloseddefault54.3%not datedLiveBench Leaderboard
25gemini-2.0-flash-liteGoogle DeepMindcloseddefault54.3%not datedLiveBench Leaderboard
26gemini-2.0-flash-lite-preview-02-05Google DeepMindcloseddefault53.2%not datedLiveBench Leaderboard
27Dracarys2-72B-Instructnot stateddefault52.6%not datedLiveBench Leaderboard
28gpt-4o-2024-11-20OpenAIcloseddefault52.2%not datedLiveBench Leaderboard
29learnlm-1.5-pro-experimentalnot stateddefault52.2%not datedLiveBench Leaderboard
30meta-llama/Llama-3.3-70B-InstructEpoch: Llama-3.3-70B-Instruct, Meta AIopendefault50.2%not datedLiveBench Leaderboard
31google/gemma-3-27b-itEpoch: gemma-3-27b-it, Google DeepMindopendefault50.0%not datedLiveBench Leaderboard
32claude-3-opus-20240229Anthropiccloseddefault49.2%not datedLiveBench Leaderboard
33mistral-large-2411Mistral AIno page matchopendefault48.4%not datedLiveBench Leaderboard
34sonarnot stateddefault46.9%not datedLiveBench Leaderboard
35Qwen/Qwen2.5-Coder-32B-InstructEpoch: Qwen2.5-Coder-32B-Instruct, Alibabaopendefault46.2%not datedLiveBench Leaderboard
36Dracarys2-Llama-3.1-70B-Instructnot stateddefault46.2%not datedLiveBench Leaderboard
37deepseek-ai/DeepSeek-R1-Distill-Qwen-32BEpoch: DeepSeek-R1-Distill-Qwen-32B, DeepSeekopendefault45.6%not datedLiveBench Leaderboard
38mistral-small-2503Mistral AIno page matchopendefault44.0%not datedLiveBench Leaderboard
39amazon.nova-pro-v1:0Amazoncloseddefault43.5%not datedLiveBench Leaderboard
40claude-3-5-haiku-20241022Anthropiccloseddefault43.5%not datedLiveBench Leaderboard
41mistral-small-2501Mistral AIno page matchopendefault42.5%not datedLiveBench Leaderboard
42microsoft/phi-4Epoch: phi-4, Microsoft Researchopendefault41.6%not datedLiveBench Leaderboard
43gpt-4o-mini-2024-07-18OpenAIcloseddefault41.3%not datedLiveBench Leaderboard
44Qwen/QwQ-32B-PreviewEpoch: QwQ-32B-Preview, Alibabaopendefault40.3%not datedLiveBench Leaderboard
45google/gemma-2-27b-itEpoch: gemma-2-27b-it, Google DeepMindopendefault38.2%not datedLiveBench Leaderboard
46amazon.nova-lite-v1:0Amazoncloseddefault36.4%not datedLiveBench Leaderboard
47CohereLabs/c4ai-command-r-plus-08-2024Epoch: c4ai-command-r-plus-08-2024, Cohere,Cohere Labs (formerly Cohere for AI)opendefault31.8%not datedLiveBench Leaderboard
48amazon.nova-micro-v1:0Amazoncloseddefault29.6%not datedLiveBench Leaderboard
49google/gemma-2-9b-itEpoch: gemma-2-9b-it, Google DeepMindopendefault28.7%not datedLiveBench Leaderboard
50CohereLabs/c4ai-command-r-08-2024Epoch: c4ai-command-r-08-2024opendefault27.5%not datedLiveBench Leaderboard
51microsoft/Phi-3-small-8k-instructEpoch: Phi-3-small-8k-instruct, Microsoftopendefault24.0%not datedLiveBench Leaderboard
52microsoft/Phi-3-mini-4k-instructEpoch: Phi-3-mini-4k-instruct, Microsoftopendefault22.4%not datedLiveBench Leaderboard
53allenai/OLMo-2-1124-13B-InstructEpoch: OLMo-2-1124-13B-Instruct, Allen Institute for AI,University of Washington,New York University (NYU)opendefault22.1%not datedLiveBench Leaderboard
54microsoft/Phi-3-mini-4k-instructEpoch: Phi-3-mini-4k-instruct, Microsoftopendefault22.1%not datedLiveBench Leaderboard

Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method