Knowledge & reasoning

BBH leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-10, the highest measured score on BBH is 78.8% by Soofi S 30B-A3B. 110 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Celeris-1	100.0%	estimated ± 3.2 pp, low confidence
2	Claude Opus 4.5	100.0%	estimated ± 3.2 pp, low confidence
3	Claude Opus 4.6	100.0%	estimated ± 3.2 pp, low confidence
4	Claude Sonnet 4.6	100.0%	estimated ± 3.2 pp, low confidence
5	DeepSeek-R1	100.0%	estimated ± 3.2 pp, low confidence
6	DeepSeek V3	100.0%	estimated ± 3.2 pp, low confidence
7	DeepSeek V3.1	100.0%	estimated ± 3.2 pp, low confidence
8	DeepSeek V3.1 (Reasoning)	100.0%	estimated ± 3.2 pp, low confidence
9	DeepSeek V4 Flash 0731	100.0%	estimated ± 3.2 pp, low confidence
10	DeepSeek V4 Pro 0813	100.0%	estimated ± 3.2 pp, low confidence
11	Exaone 4.0 32B	100.0%	estimated ± 3.2 pp, low confidence
12	Gemma 4 12B	100.0%	estimated ± 3.2 pp, low confidence
13	Gemma 4 26B A4B	100.0%	estimated ± 3.2 pp, low confidence
14	Gemma 4 31B	100.0%	estimated ± 3.2 pp, low confidence
15	GLM-4.7	100.0%	estimated ± 3.2 pp, low confidence
16	GLM-5	100.0%	estimated ± 3.2 pp, low confidence
17	Granite 4.2 30B	100.0%	estimated ± 3.2 pp, low confidence
18	Granite 4.2 8B	100.0%	estimated ± 3.2 pp, low confidence
19	Grok 3 [Beta]	100.0%	estimated ± 3.2 pp, low confidence
20	K-EXAONE 2.0	100.0%	estimated ± 3.2 pp, low confidence
21	Kimi K2	100.0%	estimated ± 3.2 pp, low confidence
22	Kimi K2.5	100.0%	estimated ± 3.2 pp, low confidence
23	Kimi K2.5 (Reasoning)	100.0%	estimated ± 3.2 pp, low confidence
24	LongCat-Flash-Lite-Sparse	100.0%	estimated ± 3.2 pp, low confidence
25	MAI-Thinking-1	100.0%	estimated ± 3.2 pp, low confidence
26	MiMo-V2-Flash	100.0%	estimated ± 3.2 pp, low confidence
27	MiniCPM5-2B	100.0%	estimated ± 3.2 pp, low confidence
28	Nemotron 3.5 Lightning 30B A3B NVFP4	100.0%	estimated ± 3.2 pp, low confidence
29	Nemotron 3 Nano 30B	100.0%	estimated ± 3.2 pp, low confidence
30	Nemotron 3 Nano Omni 30B A3B	100.0%	estimated ± 3.2 pp, low confidence
31	Nemotron 3 Super 100B	100.0%	estimated ± 3.2 pp, low confidence
32	Nemotron 3 Ultra	100.0%	estimated ± 3.2 pp, low confidence
33	Qwen3 235B 2507	100.0%	estimated ± 3.2 pp, low confidence
34	Qwen3 235B 2507 (Reasoning)	100.0%	estimated ± 3.2 pp, low confidence
35	Qwen3.5-122B-A10B	100.0%	estimated ± 3.2 pp, low confidence
36	Qwen3.5-27B	100.0%	estimated ± 3.2 pp, low confidence
37	Qwen3.5-35B-A3B	100.0%	estimated ± 3.2 pp, low confidence
38	Qwen3.5 397B	100.0%	estimated ± 3.2 pp, low confidence
39	Qwen3.6-27B	100.0%	estimated ± 3.2 pp, low confidence
40	Qwen3.6-35B-A3B	100.0%	estimated ± 3.2 pp, low confidence
41	Qwen3.6 Plus	100.0%	estimated ± 3.2 pp, low confidence
42	Qwen3.7 Max	100.0%	estimated ± 3.2 pp, low confidence
43	Qwen3.7 Plus	100.0%	estimated ± 3.2 pp, low confidence
44	Sarvam 105B	100.0%	estimated ± 3.2 pp, low confidence
45	Sarvam 30B	100.0%	estimated ± 3.2 pp, low confidence
46	Solar Open 2	100.0%	estimated ± 3.2 pp, low confidence
47	Solar Pro 4	100.0%	estimated ± 3.2 pp, low confidence
48	ZAYA1-8B	100.0%	estimated ± 3.2 pp, low confidence
49	Gemma 4 E4B	98.9%	estimated ± 3.2 pp, low confidence
50	Qwen2.5-VL-32B	98.2%	estimated ± 3.2 pp, low confidence
51	ZAYA1-74B-Preview	97.4%	estimated ± 3.2 pp, low confidence
52	Granite 4.2 3B	97.1%	estimated ± 3.2 pp, low confidence
53	Gemma 4 E2B	88.1%	estimated ± 3.2 pp, low confidence
54	Soofi S 30B-A3B	78.8%	measured
55	Claude Mythos 5	78.7%	estimated ± 10.9 pp, low confidence
56	Claude Opus 4.7 (Adaptive)	78.7%	estimated ± 10.9 pp, low confidence
57	Claude Opus 4.8	78.7%	estimated ± 10.9 pp, low confidence
58	Claude Sonnet 4.5	78.7%	estimated ± 10.9 pp, low confidence
59	DeepSeek V4.1 Flash	78.7%	estimated ± 10.9 pp, low confidence
60	Gemini 2.5 Pro	78.7%	estimated ± 10.9 pp, low confidence
61	Gemini 3.5 Flash	78.7%	estimated ± 10.9 pp, low confidence
62	GLM-4.7-Flash	78.7%	estimated ± 10.9 pp, low confidence
63	GLM-5.2	78.7%	estimated ± 10.9 pp, low confidence
64	GPT-4.1	78.7%	estimated ± 10.9 pp, low confidence
65	GPT-5.2	78.7%	estimated ± 10.9 pp, low confidence
66	GPT-5.4	78.7%	estimated ± 10.9 pp, low confidence
67	GPT-5.4 mini	78.7%	estimated ± 10.9 pp, low confidence
68	GPT-5.4 nano	78.7%	estimated ± 10.9 pp, low confidence
69	GPT-5.5	78.7%	estimated ± 10.9 pp, low confidence
70	GPT-5.6 Luna	78.7%	estimated ± 10.9 pp, low confidence
71	GPT-5.6 Sol	78.7%	estimated ± 10.9 pp, low confidence
72	GPT-5.6 Terra	78.7%	estimated ± 10.9 pp, low confidence
73	GPT-6 Astra	78.7%	estimated ± 10.9 pp, low confidence
74	GPT-OSS 120B	78.7%	estimated ± 10.9 pp, low confidence
75	GPT-OSS 20B	78.7%	estimated ± 10.9 pp, low confidence
76	Grok 3 Mini	78.7%	estimated ± 10.9 pp, low confidence
77	Grok 4.3	78.7%	estimated ± 10.9 pp, low confidence
78	Hy3 Preview	78.7%	estimated ± 10.9 pp, low confidence
79	Hy4 preview	78.7%	estimated ± 10.9 pp, low confidence
80	Inkling	78.7%	estimated ± 10.9 pp, low confidence
81	Inkling-Small	78.7%	estimated ± 10.9 pp, low confidence
82	Interfaze Beta	78.7%	estimated ± 10.9 pp, low confidence
83	Kimi K2.6	78.7%	estimated ± 10.9 pp, low confidence
84	Kimi K3	78.7%	estimated ± 10.9 pp, low confidence
85	Ling 3.0 Flash	78.7%	estimated ± 10.9 pp, low confidence
86	Ling 3.0 Flash FP8	78.7%	estimated ± 10.9 pp, low confidence
87	Ministral 3 14B (Reasoning)	78.7%	estimated ± 10.9 pp, low confidence
88	Nemotron Ultra 253B	78.7%	estimated ± 10.9 pp, low confidence
89	o1	78.7%	estimated ± 10.9 pp, low confidence
90	o1-pro	78.7%	estimated ± 10.9 pp, low confidence
91	o3-mini	78.7%	estimated ± 10.9 pp, low confidence
92	Ornith-1.5-35B-A3B	78.7%	estimated ± 10.9 pp, low confidence
93	Ornith-1.5-397B	78.7%	estimated ± 10.9 pp, low confidence
94	Ornith-1.5-9B	78.7%	estimated ± 10.9 pp, low confidence
95	Qwen3.8-27B	78.7%	estimated ± 10.9 pp, low confidence
96	Qwen3.8-Flash-Next	78.7%	estimated ± 10.9 pp, low confidence
97	Qwen3.8 Max	78.7%	estimated ± 10.9 pp, low confidence
98	Qwen3.8-Omni-Flash	78.7%	estimated ± 10.9 pp, low confidence
99	Sakana Fugu	78.7%	estimated ± 10.9 pp, low confidence
100	Sakana Fugu-Ultra	78.7%	estimated ± 10.9 pp, low confidence
101	Ternary Bonsai 2 27B	78.7%	estimated ± 10.9 pp, low confidence
102	GPT-4.1 mini	78.7%	estimated ± 10.9 pp, low confidence
103	Claude 3.5 Sonnet	78.7%	estimated ± 10.9 pp, low confidence
104	Ling 2.6 Flash	78.7%	estimated ± 10.9 pp, low confidence
105	Mellum2-12B-A2.5B-Thinking	78.7%	estimated ± 10.9 pp, low confidence
106	Phi-4	78.7%	estimated ± 10.9 pp, low confidence
107	GPT-4.1 nano	78.7%	estimated ± 10.9 pp, low confidence
108	Mellum2-12B-A2.5B-Instruct	78.7%	estimated ± 10.9 pp, low confidence
109	MiniCPM5-1B	71.9%	measured
110	Granite-4.0-1B	60.4%	measured
111	Granite-4.0-H-1B	59.7%	measured
112	LFM2.5-230M	42.3%	estimated ± 3.2 pp, medium confidence
113	LFM2.5-350M	42.1%	estimated ± 3.2 pp, medium confidence
114	LFM2.5-VL-450M	41.3%	estimated ± 3.2 pp, medium confidence
115	Granite-4.0-H-350M	33.3%	measured
116	Granite-4.0-350M	33.1%	measured

BBH leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General