MiniMax H3 Max is fal's post-trained variant of MiniMax H3, optimized with fal's inference stack for higher throughput and offered through text, image, and reference video endpoints.
Status
active
Developer
fal
Released
Evidence
5 first-party records
Record published
Last evidence review
Identity
Model, family, and access are separate
A hosted endpoint can have different limits, timing, and pricing from its underlying model family.
Canonical name
MiniMax H3 Max
Aliases
H3 Max, fal H3 Max
Developer
fal
Available through
fal
Family relationship
Post-trained variant; see the related base-model record
Reference input limits and combined token billing remain endpoint-specific and are not flattened into this model-level claim.
Performance
Reported speed, with its boundary
under 3 seconds
fal reported this wall time for a five-second clip under its launch evaluation conditions. H3 Max Stream has not independently reproduced the benchmark.
Scope: Five-second clip under fal's launch evaluation conditions; methodology is provider-reported and not independently reproduced by H3 Max Stream.