MiniMax launches H3 Image: text-to-image and editing with up to 16 references and ~20MP output, about $0.022 per 1K image; billed as open-weight

Why it matters

【Key points】

• · Nov 10, 2026: MiniMax adds MiniMax H3 Image to its API platform.
• MiniMax calls it "a leading open-weight image model".
• A faster H3-Image-Highspeed variant ships alongside.

• · One model covers text-to-image and editing, with 1–16 reference images.
• Output goes up to ~20MP (4K tier = 4096×4096), at aspect ratios from 1:3 to 3:1.
• Each request returns 1–10 images, and results can be streamed.

• · Standard pricing per 1M tokens: text input $2.30, image input $1.50, image output $3.80.
• That works out to about $0.022 per 1K image, $0.023 per 2K image and $0.069 per 4K image.
• 4K is 50% off (~$0.034) until Nov 17 16:00 UTC.

· The Highspeed variant costs $7.60 per 1M output tokens, about $0.035 per 1K image.

· No public Hugging Face repo was available at the time of writing.

【Our read】

· Background: this is MiniMax's first image model in the H line. H3 (Jul 31) was an open-weight video model. image-01 has been moved to the legacy list.

• · Comparison: OpenAI Images 2.6 (e/1088) and Canva Craft (e/1044) are closed.
• FLUX 3 Image Klein (e/889) has ~5.6B open weights and takes up to 10 references.
• Bria Fibo 2.0 (e/994) is non-commercial only.
• H3 Image's specs are more aggressive, but MiniMax has published no benchmark scores yet.

· Impact: at ~2¢ per image, H3 Image sits in the low-price tier of mainstream image APIs. Two things will decide its impact: whether the weights ship with a commercial license, and how it ranks on third-party leaderboards.

【Verifiable】

• ·
• ·
• ·
• ·