// BENCHMARK

LiveCodeBench benchmark

AI model leaderboard for the LiveCodeBench benchmark. Compare how large language models score on LiveCodeBench, see the full ranking, and understand what this AI benchmark measures. Gemini 3 Pro currently leads with 90.7%. Contamination-resistant coding benchmark continuously updated from new competitive-programming problems.

Leaderboard

# Model Organization Score Variant Source
#1 Gemini 3 Pro Google DeepMind 90.7% cited-deepseek-v3-2 official ↗
#2 NVIDIA Nemotron 3 Ultra NVIDIA 89% official ↗
#3 MAI-Thinking-1 Microsoft 87.7% official ↗
#4 GPT-5 OpenAI 84.5% cited-deepseek-v3-2 official ↗
#5 DeepSeek-V3.2 DeepSeek 83.3% thinking official ↗
#6 MiniMax-M2 MiniMax 83% cited-deepseek-v3-2 official ↗
#7 Kimi K2 Thinking Moonshot 82.6% thinking-cited-deepseek-v3-2 official ↗
#8 Grok 4 xAI 81.9% 2407-2501 official ↗
#9 gemini-2.5-pro Google DeepMind 80.1% 2407-2501 official ↗
#10 DeepSeek V4-Pro DeepSeek 78.6% official ↗
#11 o3 OpenAI 78.4% 2407-2501 official ↗
#12 Qwen3-235B-A22B-Thinking (Jul 2025) Qwen 78.2% 2407-2501 official ↗
#13 DeepSeek-R1-0528 DeepSeek 77% 2407-2501 official ↗
#14 DeepSeek-V3.2-Exp DeepSeek 74.1% from-hf-readme official ↗
#15 Llama Nemotron Super 49B NVIDIA 73.58% reasoning-on-lcb-oct24-feb25 official ↗
#16 GLM 4.5 Zhipu AI 72.9% 2407-2501 official ↗
#17 Gemma 4 12B Google DeepMind 72% official ↗
#18 NVIDIA-Nemotron-Nano-9B-v2 NVIDIA 71.1% from-hf-readme official ↗
#19 NVIDIA-Nemotron-Nano-12B-v2 NVIDIA 70.79% from-hf-readme official ↗
#20 GLM-4.5-Air Zhipu AI 70.7% 2407-2501 official ↗
#21 Llama Nemotron Ultra 253B NVIDIA 66.31% reasoning-on-lcb-aug24-feb25 official ↗
#22 MiniMax-M1-80k MiniMax 65% 16 samples official ↗
#23 Claude Sonnet 4.5 Anthropic 64% cited-deepseek-v3-2 official ↗
#24 Claude Opus 4 Anthropic 63.6% 2407-2501 official ↗
#25 MiniMax-M1-40k MiniMax 62.3% 16 samples official ↗
#26 Ling-1T Ant Group 61.68% pass@1 official ↗
#27 Pangu Pro MoE Huawei 59.6% instruct-pass@1 official ↗
#28 EXAONE Deep 32B LG AI Research 59.5% pass@1-from-hf-readme official ↗
#29 GPT-5 OpenAI 58.57% main-cited-ling-1t official ↗
#30 DeepSeek-R1-Distill-Llama-70B DeepSeek 57.5% pass@1-cited-exaone official ↗
#31 DeepSeek-R1-Distill-Qwen-32B Qwen 57.2% pass@1-cited-exaone official ↗
#32 DeepSeek-R1-Distill-Qwen-32B DeepSeek 57.2% pass@1-cited-exaone official ↗
#33 EXAONE Deep 7.8B LG AI Research 55.2% pass@1-from-hf-readme official ↗
#34 Kimi K2 Moonshot 48.95% instruct-0905-cited-ling-1t official ↗
#35 DeepSeek-V3.1-Terminus DeepSeek 48.02% non-thinking-cited-ling-1t official ↗
#36 LongCat-Flash Meituan Inc 48.02% pass@1 official ↗
#37 EXAONE Deep 2.4B LG AI Research 46.6% pass@1-from-hf-readme official ↗
#38 gemini-2.5-pro Google DeepMind 45.43% lowthink-cited-ling-1t official ↗
#39 DeepSeek-V3 DeepSeek 42.7% pass@1 official ↗
#40 DeepSeek-V2.5 DeepSeek 41.8% jan-sep-2024-from-hf-readme official ↗
#41 DeepSeek-R1-Distill-Llama-8B DeepSeek 39.6% pass@1-cited-exaone official ↗
#42 Llama 4 Scout Meta AI 33.1% pass@1-cited-pangu-pro-moe official ↗
#43 dots.llm1 Rednote 32% pass@1 official ↗
#44 Llama Nemotron Ultra 253B NVIDIA 29.03% reasoning-off-lcb-aug24-feb25 official ↗
#45 DeepSeek-R1-Distill-Qwen-1.5B Qwen 16.9% pass@1-cited-exaone official ↗
#46 DeepSeek-R1-Distill-Qwen-1.5B DeepSeek 16.9% pass@1-cited-exaone official ↗

Frequently asked questions about LiveCodeBench

What is the LiveCodeBench benchmark?

Contamination-resistant coding benchmark continuously updated from new competitive-programming problems.

How is the LiveCodeBench benchmark scored?

LiveCodeBench is scored using the pass@1 (%) metric, where a higher score is better. The maximum achievable score is 100.000. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on LiveCodeBench?

As of the latest reported scores on GenAIList, Gemini 3 Pro achieves the highest result on LiveCodeBench with a score of 90.7%.

Is a higher LiveCodeBench score better?

Yes. On LiveCodeBench a higher score indicates better performance, so models near the top of the leaderboard are the strongest.