Vision & documents

V* leaderboard

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

As of 2026-10-07, the highest measured score on V* is 96.9% by Kimi K2.6. 16 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Kimi K3	100.0%	estimated ± 5.1 pp, low confidence
2	Qwen3.8 Max	100.0%	estimated ± 5.1 pp, low confidence
3	Seed 2.1 Pro	99.7%	estimated ± 5.1 pp, low confidence
4	Qwen3.8-Omni-Flash	98.8%	estimated ± 5.1 pp, low confidence
5	Qwen3.8-Flash-Next	97.4%	estimated ± 5.1 pp, low confidence
6	Qwen3.7 Plus	97.0%	estimated ± 5.1 pp, low confidence
7	Kimi K2.6	96.9%	measured
8	Qwen3.6 Plus	96.9%	measured
9	Seed 2.1 Turbo	96.8%	estimated ± 5.1 pp, low confidence
10	Qwen3.8-27B	96.7%	estimated ± 5.1 pp, low confidence
11	Qwen3.5 397B	95.8%	measured
12	Step 3.7 Flash	95.3%	measured
13	Qwen3.6-27B	94.7%	measured
14	Qwen3.5-27B	93.7%	measured
15	dots3-note Preview	93.6%	estimated ± 5.1 pp, medium confidence
16	Qwen3.6-35B-A3B	93.3%	estimated ± 1.4 pp, medium confidence
17	Qwen3.5-122B-A10B	93.2%	measured
18	Qwen3.5-35B-A3B	92.7%	measured
19	Command A+	89.1%	estimated ± 1.4 pp, low confidence
20	Gemini 3 Pro	88.0%	measured
21	Nemotron 3 Nano Omni 30B A3B	86.2%	estimated ± 1.4 pp, low confidence
22	Gemma 4 12B	80.0%	estimated ± 5.1 pp, medium confidence
23	GPT-5.2	75.9%	measured
24	LFM2.5-VL-3B	68.2%	estimated ± 1.4 pp, low confidence
25	Claude Opus 4.5	67.0%	measured
26	ZAYA1-VL-8B	66.0%	estimated ± 1.4 pp, low confidence
27	LFM2.5-VL-450M	51.8%	estimated ± 1.4 pp, low confidence

V* leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General