LiveCodeBench - February to May 2025

Code

LiveCodeBench code-generation problems released between February 1 and May 1, 2025, the window Qwen labels "LiveCodeBench v6 (25.02-25.05)". This exact historical problem cohort is retained as a diagnostic rather than pooled with other LiveCodeBench windows.

Top models (higher is better)

Model	Score
Qwen3 235B A22B Thinking 2507	74.1
Gemini 2.5 Pro	72.5
O4 Mini	71.8
Qwen3 VL 235B A22B Thinking	70.1
Qwen3-Max (2025-09-23)	69.0
DeepSeek-R1-0528	68.7
Qwen3-Next-80B-A3B-Thinking	68.7
Qwen3 30B A3B Thinking 2507	66.0
Qwen3 VL 32B Thinking	65.6
Qwen3 VL 30B A3B Thinking	64.2
Gemini 2.5 Flash	61.2
Qwen3 32B	60.6
o3	58.6
Qwen3 VL 8B Thinking	58.6
Qwen3-30B-A3B	57.4
Qwen3 Next 80B A3B Instruct	56.6
Qwen3-235B-A22B	55.7
Qwen3 4B Thinking 2507	55.2
Qwen3-VL-235B-A22B-Instruct	54.3
DeepSeek-V3.1	52.3
Qwen3 235B A22B Instruct 2507	51.8
Qwen3 VL 4B Thinking	51.3
Qwen3 8B	51.0
Opus 4	48.9
Kimi K2 Instruct	48.9
Qwen3 4B	48.4
DeepSeek-V3-0324	45.2
Qwen3 VL 32B Instruct	43.8
Qwen3 30B A3B Instruct 2507	43.2
Qwen3 VL 30B A3B Instruct	42.6
Qwen3-VL-8B-Instruct	39.3
Qwen3 VL 4B Instruct	37.9
ChatGPT-4o Latest observed 2025-03-27	35.8
Qwen3 4B Instruct 2507	35.1
GPT-4.1 Nano	31.5
Qwen3 1.7B	31.3
Qwen3 VL 2B Thinking	29.3
Qwen3 VL 2B Instruct	20.3