Size
33.1B
a dense model — all of it works on every answer
MiniMax H3-Image
Context
Not yet confirmed
the longest conversation the model is built to hold
Lab
MiniMax
the lab that released it
Released
Nov 10, 2026
Licence: MiniMax-H3-Image Community License Agreement — the licence covers use ONLY outside its Excluded Territories: the European Union, the United Kingdom, South Korea and the United States are excluded, and use there is not authorised. Separate prior written authorisation from MiniMax is required if your annualised revenue exceeds $1M or you provide the model as a hosted service; commercial products must display 'MiniMax-H3-Image'; outputs may not be used to improve another AI model. Read it before deploying — for most Western renters the territory clause ends the conversation.
Worth knowing
MiniMax's open image generation and editing model, released Nov 10, 2026: a 33.1B dense diffusion transformer that generates up to 16-megapixel images (4096x4096) natively in 6 sampling steps and edits with up to 16 reference images. The BF16 checkpoint is 68.65GB (transformer 66.28GB + VAE 2.34GB) and serves on one GPU via vLLM-Omni or SGLang — MiniMax's card lists the H200, RTX PRO 6000 Blackwell and AMD MI350X. Two watch-outs: the checkpoint ships WITHOUT its text encoder, which loads separately from the MiniMax-H3 repository (66.7GB), and the licence excludes the EU, UK, South Korea and the US entirely — see the licence line. An optional prompt-expander companion, MiniMax-H3-Image-PE (27B, NVFP4, 17.9GB), ships under the same licence.
What it needs
Full precision (FP16)
66 GB
the weights, uncompressed
8-bit
33 GB
the weights, compressed to 8-bit
4-bit
17 GB
the weights, compressed to 4-bit
How we work this out: memory = the model's total parameters × bytes per weight — 2 bytes at full precision (FP16), 1 byte at 8-bit, half a byte at 4-bit. Weights only — add room for the conversation memory (KV cache) and the serving software. A GPU counts as running a build when its memory covers the weights plus about 20% headroom. Real figures move with context length and software.
GPUs that can run it
Ranked by how well the model runs, then by price.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 95GB, with room left for the conversation memory.
This is the Max-Q version: it runs at lower power and is slower than the full card.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 80GB, with room left for the conversation memory.
An older generation — no FP8, so the newer speed tricks for serving don't apply.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 96GB, with room left for the conversation memory.
No NVLink scale-out; strong all-round, but the Max-Q version (listed separately) runs slower.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 80GB, with room left for the conversation memory.
No NVLink, and lower power than the SXM — slower when cards must work together.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 80GB, with room left for the conversation memory.
Pays off only when the card stays busy; on bursty work you pay H100 rent for idle silicon.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 94GB, with room left for the conversation memory.
Sold as linked pairs, not as a lone card.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 80GB, with room left for the conversation memory.
One provider's label, published alongside its plainer “H100 PCIe” — check which card you are actually booking.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 141GB, with room left for the conversation memory.
More card than most jobs need; the main trap is paying for it on small work.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 80GB, with room left for the conversation memory.
Pays off only when kept busy — avoid it for bursty development work.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 192GB, with room left for the conversation memory.
Frontier training money. If your job fits on anything cheaper, a B200 is budget burned for a spec sheet.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 96GB, with room left for the conversation memory.
Its processor is ARM, not the usual x86 — some software needs changes to run on it.
Runs MiniMax H3-Image at full precision — its 66GB of weights fit this card's 288GB, with room left for the conversation memory.
Very costly for small jobs — best value only on big work.
Runs MiniMax H3-Image compressed — the 8-bit build needs 33GB, which fits this card's 48GB. The compromise: slightly lower answer quality and a shorter conversation memory than the full-precision build.
An older generation — no FP8, so the newer speed tricks for serving don't apply.
Runs MiniMax H3-Image compressed — the 8-bit build needs 33GB, which fits this card's 48GB. The compromise: slightly lower answer quality and a shorter conversation memory than the full-precision build.
70B at FP16 does not fit, and no flag fixes that. Weak at double-precision maths; no NVLink.
Runs MiniMax H3-Image compressed — the 8-bit build needs 33GB, which fits this card's 48GB. The compromise: slightly lower answer quality and a shorter conversation memory than the full-precision build.
Weak at double-precision maths — not the card for scientific computing. No NVLink.
Runs MiniMax H3-Image compressed — the 8-bit build needs 33GB, which fits this card's 40GB. The compromise: slightly lower answer quality and a shorter conversation memory than the full-precision build.
The 40GB fills up quickly on larger models.
Runs MiniMax H3-Image compressed — the 4-bit build needs 17GB, which fits this card's 24GB. The compromise: slightly lower answer quality and a shorter conversation memory than the full-precision build.
24GB fills up fast on big models. No NVLink, so two cards can't pool memory — and it is a gaming card at heart.
Runs MiniMax H3-Image compressed — the 4-bit build needs 17GB, which fits this card's 32GB. The compromise: slightly lower answer quality and a shorter conversation memory than the full-precision build.
No NVLink, so two cards can't pool memory. A consumer card — driver and ECC behaviour matter for production serving.
The cheapest verified way to run it
RTX 4090 at RunPod Community — $0.34/hr — runs it compressed — the 4-bit build.
Launch RunPod Community ↗ See the RTX 4090 →
Checked Nov 15, 2026
Select a GPU See these GPUs on the prices board, ranked for this model.