Competitive coding
LiveCodeBench
Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better.
Rankings
Higher is better#193.5%
DeepSeek-V4-Provia BenchLM
via BenchLM
291.9%
Qwen3.8-Flash-NextOW
391.6%
DeepSeek-V4-Flashvia BenchLM
via BenchLM
391.6%
Qwen3.7-Maxvia BenchLM
via BenchLM
590.3%
Qwen3.8-27BOW
689.6%
Kimi K2.6via BenchLM
via BenchLM
689.6%
Qwen3.7-Plusvia BenchLM
via BenchLM
885%
Kimi K2.5via BenchLM
via BenchLM
984.9%
GLM-4.7via BenchLM
via BenchLM
1071.1%
Nemotron Nano 2OW
1168.3%
Nemotron 3 NanoOS
1266.3%
Llama Nemotron Ultra 253BOW
Scores marked “via” above are quoted with attribution from BenchLM, retrieved 24 August 2026.
LiveCodeBench — frequently asked questions
Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better.