KVBench (Brief Caption) - Math (English): leaderboard

KVBench (2026) tests whether text-to-image models can draw correct senior high-school math content, such as diagrams and scientific illustrations, from English prompts given as a brief caption, so the model must supply the textbook knowledge itself. Qwen2.5-VL-32B checks each image against a list of textbook-grounded constraints; the score is the percentage satisfied, higher is better.

Metric: Checklist score (%) on the math prompts in English, each given as a brief caption, so the model must supply the textbook knowledge itself: KVBench knowledge-intensive text-to-image prompts from over 30 senior high-school textbooks (150 per subject and language, each with a textbook reference image); Qwen2.5-VL-32B-Instruct answers a checklist of four to six binary questions (key objects, attributes, spatial relations, reasoning outcomes) about each generated image, and the score is the mean share of satisfied items, in percent; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Seedream 4.040.38
2GPT-Image (KVBench checkpoint unspecified)38.04
3Nano Banana Pro (Gemini 3 Pro Image)37.07
4FLUX.2-dev35.42
5FLUX 2 Max34.6
6FLUX.1-dev34.22
7SD3.5-Large (KVBench GGUF build, quantization unspecified)28.84
8FLUX.1-Krea-dev27.58
9Qwen-Image27.16
10Show-o226.22
11Z-Image25.64
12OmniGen224.62#4579
13Janus-Pro23.38
14BAGEL-7B-MoT21.98#4619

Interactive version: theaggregate.ai/benchmark?slug=kvbench-brief-caption-math-english · How It Works · Data refreshed daily, snapshot 2026-11-21.