LiveCodeBench

一项无污染的编程基准，持续从 LeetCode、AtCoder 和 CodeForces 采集新的竞赛编程题目，评估代码生成、自我修复和代码执行。

分数分布

包括已弃用的模型

排名	模型	评分
51	Ministral 3 14B Mistral	35.1%
52	Nova 2.0 Lite (Non-reasoning) Amazon	34.6%
53	NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning) NVIDIA	34.5%
54	Nova Premier Amazon	31.7%
55	Llama 3.1 Instruct 405B Meta	30.5%
56	Nova 2.0 Omni (Non-reasoning) Amazon	30.5%
57	Ministral 3 8B Mistral	30.3%
58	Llama 4 Scout Meta	29.9%
59	Exaone 4.0 1.2B (Non-reasoning) LG AI Research	29.3%
60	Llama 3.3 Instruct 70B Meta	28.8%
61	Command A Cohere	28.7%
62	Hermes 4 - Llama-3.1 70B (Non-reasoning) Nous Research	26.9%
63	Olmo 3 7B Instruct Allen Institute for AI	26.6%
64	Granite 4.0 H Small IBM	25.1%
65	Ministral 3 3B Mistral	24.7%
66	Phi-4 Microsoft	23.1%
67	Llama 3.2 Instruct 90B (Vision) Meta	21.4%
68	Jamba Reasoning 3B AI21 Labs	21.0%
69	DeepHermes 3 - Mistral 24B Preview (Non-reasoning) Nous Research	19.5%
70	Jamba 1.7 Large AI21 Labs	18.1%
71	Granite 4.0 Micro IBM	18.0%
72	Nova Micro Amazon	14.0%
73	Phi-4 Multimodal Instruct Microsoft	13.1%
74	Phi-4 Mini Instruct Microsoft	12.6%
75	Granite 4.0 H 1B IBM	11.5%
76	Llama 3.2 Instruct 11B (Vision) Meta	11.0%
77	DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning) Nous Research	8.5%
78	Jamba 1.7 Mini AI21 Labs	6.1%
79	Molmo 7B-D Allen Institute for AI	3.9%
80	Granite 4.0 350M IBM	2.4%
81	Granite 4.0 H 350M IBM	1.9%
82	Gemma 3 270M Google	0.3%