ImageBench V1 —

192 evaluations across 6 categories

Benchmark V1 verdicts are produced by VLM judges and can contain mistakes. Treat PASS/FAIL labels as machine-assisted assessments, and inspect the images yourself. Learn more about the methodology.

Generation Details

Source-backed model context, size, cost, and request settings for this ImageBench V1 run.

local/minimax-h3-image-33b

Local

MiniMax-H3-Image is MiniMax's 33B open-weight text-to-image and editing model (a 33.1B dense single-stream diffusion transformer with the MiniMax-H3 Qwen3-VL text encoder), released November 2026 and guidance-distilled to 6 steps. Released under the MiniMax-H3-Image Community License, which does not license the model in the EU, UK, US or South Korea.

Reproducibility

Download on motif64.sh ↗

You can download the exact source code and configuration that was used to generate these images, packaged as a capsule: MiniMax-H3-Image 33B, 6 steps (diffusers) (NVIDIA GB10, 88 GB). Run it on your own machine with:

m64 m64_a970a7a75864

Maker: MiniMax
Family: MiniMax-H3
Model Size: 33B
high
Cost: local run; no API price
not_applicable
Run Target: gx10/minimax-h3-image-33b
Effective Request: response_format: b64_json · size: 2048x2048 · num_inference_steps: 6 · seed: 7
Sources: MiniMaxAI/MiniMax-H3-Image model card MiniMax-AI/MiniMax-H3-Image (GitHub)