Vision & documents

PerceptionBench leaderboard

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

As of 2026-11-07, the highest measured score on PerceptionBench is 68.3% by Kimi K3.1 Pro. 44 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6.1 Terra	72.9%	estimated ± 2.6 pp, low confidence
2	Kimi K3.1 Pro	68.3%	measured
3	Gemini 3.1 Pro	68.1%	estimated ± 2.6 pp, medium confidence
4	Gemini 3.5 Flash	67.2%	estimated ± 2.6 pp, medium confidence
5	Claude Mythos 5	67.2%	estimated ± 3.5 pp, medium confidence
6	GPT-5.6 Sol	65.4%	estimated ± 2.6 pp, medium confidence
7	Inkling 2 Small	64.7%	measured
8	Qwen3.8 Max	63.5%	measured
9	Gemini 3.8 Flash	61.8%	estimated ± 3.5 pp, medium confidence
10	Seed 2.1 Pro	61.3%	estimated ± 2.6 pp, medium confidence
11	GPT-5.4	60.1%	estimated ± 2.6 pp, medium confidence
12	GPT-5.5	60.1%	estimated ± 2.6 pp, medium confidence
13	Gemini 3 Pro	59.5%	estimated ± 2.6 pp, medium confidence
14	GPT-5.6 Terra	58.6%	estimated ± 2.6 pp, medium confidence
15	Kimi K3	58.5%	measured
16	Qwen3.8-Flash-Next	58.5%	estimated ± 3.5 pp, medium confidence
17	Gemini 3.7 Flash	58.3%	estimated ± 3.5 pp, medium confidence
18	Muse Spark	57.7%	estimated ± 2.6 pp, medium confidence
19	Seed 2.1 Turbo	56.8%	estimated ± 2.6 pp, medium confidence
20	Qwen3.8-27B	56.6%	estimated ± 3.5 pp, medium confidence
21	Qwen3.8-Omni-Flash	56.2%	estimated ± 3.5 pp, medium confidence
22	GPT-5.2	55.0%	estimated ± 2.6 pp, medium confidence
23	Kimi K2.6	54.7%	estimated ± 2.6 pp, medium confidence
24	Qwen3.5 397B	53.5%	estimated ± 2.6 pp, low confidence
25	Qwen3.7 Plus	53.5%	estimated ± 2.6 pp, low confidence
26	dots3-note Preview	53.4%	measured
27	Claude Opus 4.7 (Adaptive)	53.2%	estimated ± 3.5 pp, low confidence
28	Qwen3.6 Plus	52.9%	estimated ± 2.6 pp, low confidence
29	Hy4	52.3%	estimated ± 2.6 pp, low confidence
30	Kimi K2.5	52.0%	estimated ± 2.6 pp, low confidence
31	Kimi K2.5 (Reasoning)	52.0%	estimated ± 2.6 pp, low confidence
32	GPT-5.6 Luna	51.7%	estimated ± 2.6 pp, low confidence
33	Grok 4.3	50.8%	estimated ± 2.6 pp, low confidence
34	MiniMax M3	50.8%	estimated ± 2.6 pp, low confidence
35	Pareto 26.9	50.5%	estimated ± 2.6 pp, low confidence
36	MiMo-V2.5	50.2%	estimated ± 2.6 pp, low confidence
37	Claude Opus 4.8	49.8%	estimated ± 3.5 pp, low confidence
38	Claude Opus 4.6	48.4%	estimated ± 2.6 pp, low confidence
39	Gemma 4 31B	47.2%	estimated ± 2.6 pp, low confidence
40	Qwen3.5-122B-A10B	47.2%	estimated ± 2.6 pp, low confidence
41	GPT-5.4 mini	46.3%	estimated ± 2.6 pp, low confidence
42	Inkling	44.7%	estimated ± 3.5 pp, low confidence
43	Step 5 Preview	44.5%	estimated ± 2.6 pp, low confidence
44	Qwen3.6-27B	43.9%	estimated ± 2.6 pp, low confidence
45	Inkling-Small	43.2%	estimated ± 3.5 pp, low confidence
46	Qwen3.6-35B-A3B	42.4%	estimated ± 2.6 pp, low confidence
47	Grok 4.20	42.1%	estimated ± 2.6 pp, low confidence
48	Qwen3.5-35B-A3B	41.8%	estimated ± 2.6 pp, low confidence
49	Qwen3.5-27B	41.5%	estimated ± 2.6 pp, low confidence

PerceptionBench leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General