Wait Which Model?
← Back to Video models
MiniMaxNotable

MiniMax H3 Max

Released Aug 27, 2026

Speed-oriented variant of MiniMax H3, post-trained by fal on top of the open H3 weights and released jointly by MiniMax and fal; MiniMax's API exposes it as a separate model ID.

Price per second
$0.080
Max clip length
15 s
Max resolution
1080p
Native audio
Yes
AA arena Elo
1152
LMArena Elo
—
Capabilities
text to video, image to video
Licence
Proprietary
Strengths & weaknesses

Strengths

  • Generates a 5-second clip in under 3 seconds (fal-reported, ~35x the throughput of the official H3 endpoint)
  • Cheaper than H3: $0.05/s at 480P, $0.08/s at 768P on MiniMax's API
  • Text-to-video, image-to-video and reference generation

Weaknesses

  • No 2K output on MiniMax's API (480P/768P only)
  • Minimum clip length 5 s (H3 allows 4 s)

Separate model, not just a mode: MiniMax's video guide says MiniMax-H3-Max was 'jointly released by MiniMax and fal.ai', post-trained by fal on MiniMax H3 and optimised for speed; the API enum lists it next to MiniMax-H3. Release date is the fal blog post date (2026-08-27); MiniMax's docs give none. Max resolution: MiniMax API offers 480P/768P only, but fal's own endpoint also offers 1080p ($0.16/s list), hence maxResolution 1080p. fal list prices: $0.05/s 480p, $0.08/s 768p, $0.16/s 1080p (fal page showed a 40%-off promo until Oct 15). nativeAudio: fal's page says 'every generation comes back with synchronized audio'. No H3 Max weights found on Hugging Face (MiniMaxAI org lists only MiniMax-H3), and fal does not say any were released, so openWeights false. fal's claims of #1 in its own human-preference evals and on AA/Design Arena are vendor-reported.

Links

API model strings

  • minimaxMiniMax-H3-Max
  • falminimax/h3-max/text-to-video
  • falminimax/h3-max/image-to-video