Agentic · tools

PinchBench leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on PinchBench is 95.7% by Pokee-Isaac 28B. 78 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	BTL-3	100.0%	estimated ± 5.0 pp, low confidence
2	BTL-4	100.0%	estimated ± 5.0 pp, low confidence
3	Qwen3.7 Max	100.0%	estimated ± 5.0 pp, low confidence
4	Qwen3.7 Plus	100.0%	estimated ± 5.0 pp, low confidence
5	LFM3-1.7B	99.0%	estimated ± 5.0 pp, low confidence
6	Pokee-Isaac 28B	95.7%	measured
7	Mercury 2.5	95.1%	estimated ± 5.5 pp, low confidence
8	Mistral Medium 3.5 128B	95.0%	estimated ± 5.5 pp, low confidence
9	MiMo-V2.5-Pro	94.6%	estimated ± 5.5 pp, low confidence
10	Qwen3.6 Plus	94.5%	estimated ± 5.5 pp, low confidence
11	Claude Opus 4.5	94.5%	estimated ± 5.5 pp, low confidence
12	MiniMax M2.7	94.4%	estimated ± 5.5 pp, low confidence
13	Qwen3.6-35B-A3B	94.4%	estimated ± 5.5 pp, low confidence
14	Mercury 3	91.5%	estimated ± 5.5 pp, low confidence
15	Atria Dawn Preview	90.8%	estimated ± 3.6 pp, low confidence
16	GPT-5.6 Sol	90.8%	estimated ± 3.6 pp, low confidence
17	Kimi K3.1 Pro	90.8%	estimated ± 3.6 pp, low confidence
18	Ling 3.1 Flash	90.8%	estimated ± 3.6 pp, low confidence
19	GPT-6 Astra	90.7%	estimated ± 3.6 pp, low confidence
20	Kimi K3	90.7%	estimated ± 3.6 pp, low confidence
21	Claude Opus 5	90.7%	estimated ± 3.6 pp, low confidence
22	GPT-5.5 Pro	90.7%	estimated ± 3.6 pp, low confidence
23	GPT-5.4 Pro	90.7%	estimated ± 3.6 pp, low confidence
24	Step 5 Preview	90.6%	estimated ± 3.6 pp, low confidence
25	Claude Mythos 5	90.6%	estimated ± 3.6 pp, low confidence
26	GPT-5.6 Terra	90.6%	estimated ± 3.6 pp, low confidence
27	Ornith-1.5-397B	90.5%	estimated ± 3.6 pp, low confidence
28	Claude Sonnet 5	90.4%	estimated ± 3.6 pp, low confidence
29	Mistral Large 4	90.4%	estimated ± 3.6 pp, low confidence
30	GPT-5.5	90.4%	estimated ± 3.6 pp, low confidence
31	Claude Opus 4.8	90.4%	estimated ± 3.6 pp, low confidence
32	Claude Opus 4.6	90.4%	estimated ± 3.6 pp, low confidence
33	MiniMax M3	90.4%	estimated ± 3.6 pp, low confidence
34	DeepSeek V4 Pro 0813	90.4%	estimated ± 3.6 pp, low confidence
35	dots3-note Preview	90.4%	estimated ± 3.6 pp, low confidence
36	GPT-5.6 Luna	90.4%	estimated ± 3.6 pp, low confidence
37	Kimi K2.6	90.4%	estimated ± 3.6 pp, low confidence
38	Nemotron 3.5 Bolt 74B-A9B BF16	90.4%	measured
39	GPT-5.4	90.3%	estimated ± 3.6 pp, low confidence
40	Ling 3.0 Flash Sante	90.2%	estimated ± 3.6 pp, medium confidence
41	Claude Opus 4.7 (Adaptive)	90.2%	estimated ± 3.6 pp, medium confidence
42	Nemotron 3.5 Super VL 120B A12B	90.1%	estimated ± 3.6 pp, medium confidence
43	Nemotron 3.5 Super VL 120B A12B NVFP4	90.1%	estimated ± 3.6 pp, medium confidence
44	Inkling-Small	90.1%	estimated ± 3.6 pp, medium confidence
45	Inkling	90.0%	estimated ± 3.6 pp, medium confidence
46	Beam 1.0 501B-A23B	90.0%	estimated ± 3.6 pp, medium confidence
47	Beam 1.0 501B-A23B FP8	90.0%	estimated ± 3.6 pp, medium confidence
48	Nemotron 3 Ultra	90.0%	measured
49	Sage-A1.5	90.0%	measured
50	Step 3.7 Flash	90.0%	estimated ± 3.6 pp, medium confidence
51	Beam 1.0 501B-A23B NVFP4	90.0%	estimated ± 3.6 pp, medium confidence
52	Agents-A1	90.0%	estimated ± 3.6 pp, medium confidence
53	DeepSeek V4 Flash 0731	89.8%	estimated ± 3.6 pp, medium confidence
54	Ling 3.0 Flash	89.7%	estimated ± 3.6 pp, medium confidence
55	Qwen3.5 397B	89.5%	estimated ± 3.6 pp, medium confidence
56	GLM-5.1	89.5%	estimated ± 3.6 pp, medium confidence
57	Ornith-1.5-35B-A3B	89.4%	estimated ± 3.6 pp, medium confidence
58	Agents-A1-4B	89.4%	estimated ± 3.6 pp, medium confidence
59	GPT-5.2	89.3%	estimated ± 3.6 pp, medium confidence
60	Qwen3.5-122B-A10B	89.1%	estimated ± 3.6 pp, medium confidence
61	MiniCPM5-2B	89.1%	estimated ± 5.0 pp, low confidence
62	GLM-5	89.0%	estimated ± 3.6 pp, medium confidence
63	Qwen3.5-27B	88.9%	estimated ± 3.6 pp, medium confidence
64	Qwen3.5-35B-A3B	88.9%	estimated ± 3.6 pp, medium confidence
65	Kimi K2.5	88.9%	estimated ± 3.6 pp, medium confidence
66	Kimi K2.5 (Reasoning)	88.9%	estimated ± 3.6 pp, medium confidence
67	Ornith-1.5-9B	88.4%	estimated ± 3.6 pp, medium confidence
68	Spark-X2.6-4B	88.1%	estimated ± 3.6 pp, medium confidence
69	GLM-4.7	88.0%	estimated ± 3.6 pp, medium confidence
70	Nemotron 3.5 Bolt 74B-A9B	87.8%	measured
71	Sarvam 105B	87.6%	estimated ± 3.6 pp, medium confidence
72	Solar Pro 4	87.6%	estimated ± 3.6 pp, medium confidence
73	LongCat-Flash-Lite-Sparse	87.5%	estimated ± 3.6 pp, medium confidence
74	Sarvam 30B	85.1%	estimated ± 3.6 pp, low confidence
75	Nemotron 3.5 Lightning 30B A3B NVFP4	83.4%	measured
76	LFM2.5-24B-A2B	83.2%	measured
77	LLaDA2.2-flash	81.7%	measured
78	Granite 4.2 30B	79.0%	estimated ± 5.0 pp, low confidence
79	LFM2.5-2.6B	68.2%	measured
80	Granite 4.2 3B	63.3%	estimated ± 5.0 pp, low confidence
81	Granite 4.2 8B	63.3%	estimated ± 5.0 pp, low confidence
82	LLaDA2.2-mini	62.3%	measured
83	LFM2.5-8B-A1B	59.0%	estimated ± 5.0 pp, low confidence
84	Mellum2-12B-A2.5B-Thinking	52.7%	estimated ± 5.0 pp, low confidence
85	Mellum2-12B-A2.5B-Instruct	50.6%	estimated ± 5.0 pp, low confidence
86	ZAYA1-8B	43.5%	estimated ± 5.0 pp, low confidence
87	LFM2.5-VL-3B	34.6%	estimated ± 5.0 pp, low confidence
88	MiniCPM5-1B	25.7%	estimated ± 5.0 pp, low confidence

PinchBench leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General