mmlu Pro

36 models evaluated

#	Model	Provider	Type	Score
1	Qwen3.7 Max	Alibaba	Closed	89.6%
2	Claude Opus 4.5	Anthropic	Closed	89.5%
3	Qwen3.6 Plus	Alibaba	Closed	88.5%
4	Qwen3.5 397B	Alibaba	Open	87.8%
5	DeepSeek V4 Pro (Max)	DeepSeek	Open	87.5%
6	DeepSeek V4 Pro (High)	DeepSeek	Open	87.1%
7	Kimi K2.5 (Reasoning)	Moonshot AI	Closed	87.1%
8	Kimi K2.5	Moonshot AI	Open	87.1%
9	Qwen3.5-122B-A10B	Alibaba	Open	86.7%
10	DeepSeek V4 Flash (High)	DeepSeek	Open	86.4%
11	DeepSeek V4 Flash (Max)	DeepSeek	Open	86.2%
12	Qwen3.6-27B	Alibaba	Open	86.2%
13	Qwen3.5-27B	Alibaba	Open	86.1%
14	GLM-5	Z.AI	Open	85.7%
15	Qwen3.5-35B-A3B	Alibaba	Open	85.3%
16	Qwen3.6-35B-A3B	Alibaba	Open	85.2%
17	Gemma 4 31B	Google	Open	85.2%
18	MiMo-V2-Flash	Xiaomi	Open	84.9%
19	GLM-4.7	Z.AI	Open	84.3%
20	DeepSeek V4 Flash	DeepSeek	Open	83%
21	Qwen3 235B 2507	Alibaba	Open	83%
22	DeepSeek V4 Pro	DeepSeek	Open	82.9%
23	Gemma 4 26B A4B	Google	Open	82.6%
24	Claude Opus 4.6	Anthropic	Closed	82%
25	Exaone 4.0 32B	LG AI Research	Open	81.8%
26	Claude Sonnet 4.6	Anthropic	Closed	79.2%
27	Nemotron 3 Nano Omni 30B A3B	NVIDIA	Open	77.3%
28	DeepSeek V3	DeepSeek	Open	75.9%
29	ZAYA1-8B	Zyphra	Open	74.2%
30	DeepSeek V4 Pro Base	DeepSeek	Open	73.5%
31	Gemma 4 E4B	Google	Open	69.4%
32	DeepSeek V4 Flash Base	DeepSeek	Open	68.3%
33	ZAYA1-74B-Preview	Zyphra	Open	68.1%
34	Gemma 4 E2B	Google	Open	60%
35	MiniCPM5-1B	OpenBMB	Open	48.9%
36	LFM2.5-VL-450M	LiquidAI	Open	19.3%