// BENCHMARK

LiveCodeBench v5 benchmark

AI model leaderboard for the LiveCodeBench v5 benchmark. Compare how large language models score on LiveCodeBench v5, see the full ranking, and understand what this AI benchmark measures. GPT-5 currently leads with 82.68%. LiveCodeBench v5 — 599 programming problems aggregated 2024-07 to 2025-02.

Leaderboard

# Model Organization Score Variant Source
#1 GPT-5 OpenAI 82.68% High official ↗
#2 DeepSeek-R1-0528 DeepSeek 75.2% reasoning official ↗
#3 gpt-oss-120b OpenAI 74.53% official ↗
#4 o3 OpenAI 73.3% High official ↗
#5 gpt-oss-20b OpenAI 73.22% official ↗
#6 EXAONE 4.0 (32B) LG AI Research 72.6% reasoning official ↗
#7 Qwen3-235B-A22B-Thinking (Jul 2025) Qwen 70.7% reasoning official ↗
#8 Qwen3-235B-A22B Qwen 70.7% thinking-pass@1 official ↗
#9 gemini-2.5-pro Google DeepMind 69.1% cited-phi4r official ↗
#10 o3-mini OpenAI 68.8% high-cited-phi4r official ↗
#11 DeepSeek-V3.1 DeepSeek 66.59% official ↗
#12 DeepSeek-R1 DeepSeek 65.9% cited-magistral official ↗
#13 DeepSeek-R1 DeepSeek 65.9% lcb-aug24-jan25-cited-phi4r official ↗
#14 Qwen3-32B Qwen 65.7% reasoning official ↗
#15 Qwen3-32B Qwen 65.7% thinking-pass@1 official ↗
#16 QwQ-32B Qwen 65.22% official ↗
#17 K2 Think Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) 63.97% avg@16 official ↗
#18 NTele-R1-32B-V1 ZTE 63.69% lcb-aug24-feb25-from-hf-readme official ↗
#19 Qwen3-14B Qwen 63.5% thinking-pass@1 official ↗
#20 o1 OpenAI 63.4% cited-phi4r official ↗
#21 QwQ-32B Qwen 63.4% lcb-aug24-feb25-cited-phi4r official ↗
#22 Qwen3-30B-A3B Qwen 62.6% thinking-pass@1 official ↗
#23 DeepSeek-R1 DeepSeek 61.01% official ↗
#24 QwQ-32B Qwen 60.94% cited-ntele-readme official ↗
#25 EXAONE Deep 32B LG AI Research 59.5% lcb-sep24-feb25-cited-phi4r official ↗
#26 Magistral Medium 1.1 Mistral AI 59.4% official ↗
#27 gemini-2.5-pro Google DeepMind 58.24% official ↗
#28 OpenReasoning-Nemotron-32B NVIDIA 57.79% official ↗
#29 Qwen3-235B-A22B Qwen 56.64% official ↗
#30 Magistral Small 1.1 Mistral AI 55.8% official ↗
#31 Qwen3-8B Qwen 54.5% thinking-pass@1 official ↗
#32 Phi-4-Reasoning Microsoft 53.8% lcb-aug24-jan25 official ↗
#33 o1-mini OpenAI 53.8% cited-phi4r official ↗
#34 Qwen3-32B Qwen 53.24% thinking-cited-ntele-readme official ↗
#35 DeepSeek-R1-Distill-Qwen-14B Qwen 53.1% cited-mimo official ↗
#36 DeepSeek-R1-Distill-Qwen-14B DeepSeek 53.1% cited-mimo official ↗
#37 Phi-4-Reasoning-plus Microsoft 51.7% reasoning official ↗
#38 DeepSeek-R1-Distill-Qwen-32B DeepSeek 50.26% cited-ntele-readme official ↗
#39 DeepSeek-R1-Distill-Qwen-32B Qwen 50.26% cited-ntele-readme official ↗
#40 DeepSeek-R1-Zero DeepSeek 50% cited-magistral official ↗
#41 DeepSeek-V3-0324 DeepSeek 46.7% official ↗
#42 Qwen3-4B Qwen 45.5% thinking-pass@1 official ↗
#43 EXAONE 4.0 (1.2B) LG AI Research 44.6% reasoning official ↗
#44 Llama 4 Maverick Meta AI 43.4% official ↗
#45 Qwen3-30B-A3B Qwen 42.2% official ↗
#46 Claude 3.5 Sonnet Anthropic 38.9% 1022-cited-mimo official ↗
#47 DeepSeek-R1-Distill-Qwen-7B DeepSeek 37.6% cited-mimo official ↗
#48 DeepSeek-R1-Distill-Qwen-7B Qwen 37.6% cited-mimo official ↗
#49 DeepSeek-V3 DeepSeek 36.2% cited-magistral official ↗
#50 Qwen3-235B-A22B Qwen 35.3% non-thinking-pass@1 official ↗
#51 Qwen3-1.7B Qwen 33.2% thinking-pass@1 official ↗
#52 MiMo-7B-Base Xiaomi Corp 32.9% base official ↗
#53 Qwen3-32B Qwen 31.3% non-thinking-pass@1 official ↗
#54 Qwen3-30B-A3B Qwen 29.8% non-thinking-pass@1 official ↗
#55 Mistral Medium 3 Mistral AI 29.1% cited-magistral official ↗
#56 Qwen3-14B Qwen 29% non-thinking-pass@1 official ↗
#57 Gemma 3 27B Google DeepMind 27.5% official ↗
#58 Qwen3-8B Qwen 22.8% non-thinking-pass@1 official ↗
#59 Qwen3-4B Qwen 21.3% non-thinking-pass@1 official ↗
#60 Qwen3-0.6B Qwen 12.3% thinking-pass@1 official ↗
#61 Qwen3-1.7B Qwen 11.6% non-thinking-pass@1 official ↗
#62 Qwen3-1.7B Qwen 11.6% non-reasoning official ↗
#63 Qwen3-0.6B Qwen 3.6% non-thinking-pass@1 official ↗
#64 Qwen3-0.6B Qwen 3.6% non-reasoning official ↗
#65 Gemma 3 1B Google DeepMind 1.8% non-reasoning official ↗

Frequently asked questions about LiveCodeBench v5

What is the LiveCodeBench v5 benchmark?

LiveCodeBench v5 — 599 programming problems aggregated 2024-07 to 2025-02.

How is the LiveCodeBench v5 benchmark scored?

LiveCodeBench v5 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 v5?

As of the latest reported scores on GenAIList, GPT-5 achieves the highest result on LiveCodeBench v5 with a score of 82.68%.

Is a higher LiveCodeBench v5 score better?

Yes. On LiveCodeBench v5 a higher score indicates better performance, so models near the top of the leaderboard are the strongest.