The names make MiniMax H3 vs H3 Max vs H3 Max Turbo sound like three quality tiers. The real distinction is different: H3 is the open-weight foundation, while H3 Max and H3 Max Turbo are fal Research variants created through post-training and inference optimization. Each favors a different constraint—control, balanced speed and quality, or maximum iteration speed.
Want to test the difference with your own prompt? Visit the MiniMax H3 Max generator on SeeAPI or explore the wider MiniMax API experience.
How MiniMax H3, H3 Max, and H3 Max Turbo Are Related
MiniMax launched H3 as a general-purpose multimodal generation model on July 31, 2026, then released its weights on August 3. It understands combined text, image, video, and audio context and can generate clips with native stereo sound, up to 15 seconds long and up to 2K resolution.
fal Research subsequently post-trained the open model with more data and tuned it for prompt adherence and visual appeal. Its official H3 Max launch post describes a co-designed inference stack intended to preserve those quality gains at much higher throughput.
fal also describes H3 Max Turbo as a post-trained H3 variant optimized for higher throughput. It does not disclose a separate foundation model or the specific acceleration technique. Max and Turbo are therefore best treated as two fal performance profiles derived from H3—not successive MiniMax foundation-model generations.
MiniMax H3 vs H3 Max vs H3 Max Turbo at a Glance
Category | MiniMax H3 | H3 Max | H3 Max Turbo |
|---|---|---|---|
Origin | MiniMax foundation model | fal post-trained H3 variant | fal post-trained H3 variant |
Main priority | Resolution and multimodal control | Prompt adherence, aesthetics, and speed | Lowest latency and rapid iteration |
Public resolution | 768P and 2K on MiniMax API | 480P and 768P | 480P and 768P |
Duration | 4–15 seconds on MiniMax API | 5–15 seconds | Duration control; 5 seconds by default on fal |
Workflows | Text, first/last frame, multimodal references | Text and image; reference support varies by platform | Text and image, including optional end frame |
Best fit | Final delivery and reference-heavy production | Fast 768P creative production | Drafts, batches, and interactive prototypes |
Choose MiniMax H3 for Resolution and Complex References
The official MiniMax H3 API specification supports 768P or 2K output, 4–15-second clips, first- and last-frame control, and reference-to-video generation. One reference request can contain up to nine images, three video clips, and three audio clips.
That breadth matters when a scene must follow an actor image, camera-motion clip, and voice reference together. For 2K output, in-context regeneration lets H3 revisit the original multimodal context while rebuilding detail. Choose H3 when resolution, editing, voice transfer, or complex references matter more than waiting time.
Choose H3 Max for Fast, Production-Oriented 768P Video
H3 Max targets the middle ground. fal focused its post-training on instructions and aesthetics while optimizing the serving stack alongside the model. It is designed for creators who need usable 768P footage quickly.
Platform support differs. MiniMax’s direct API documents text-to-video and first/last-frame image-to-video, while fal later added a separate H3 Max reference endpoint. Check the actual service controls before building a workflow. H3 Max best fits ads, social video, product motion, and interactive tools where users wait for each result.
Choose H3 Max Turbo When Feedback Speed Drives Output
Turbo narrows the decision further. Its official fal text-to-video schema includes 480P or 768P resolution, duration, seed, safety checking, six aspect ratios, and balanced or quality prompt expansion. The image endpoint adds a starting image and optional ending image.
fal’s public example reports timings.inference of about 1.426 seconds for a five-second video. This measures backend DiT denoising, not the complete user wait; queueing, safety checks, uploads, and delivery add latency. Turbo best fits rapid A/B tests, storyboard motion drafts, batches, and latency-sensitive experiences.
MiniMax H3 vs H3 Max vs H3 Max Turbo Benchmarks
fal reports that H3 Max generated a five-second video in under three seconds, around 35 times the throughput of the official MiniMax H3 endpoint. In fal’s vendor-run human-preference evaluation against 12 video models, H3 Max ranked first for overall preference, prompt understanding, and aesthetics. The study used Bayesian Elo ratings with 95% confidence intervals.
Those results are useful launch evidence, not a universal guarantee. They were published by H3 Max’s developer. More importantly, fal has not published a separate, controlled human-preference benchmark for H3 Max Turbo. The Turbo timing example should not be combined with the Max preference study to claim that Turbo is automatically faster and better.
For a fair test, keep prompt, source frame, duration, resolution, and expansion mode constant. Record total latency, audiovisual consistency, and usable-output rate across several runs.
MiniMax H3 vs H3 Max vs H3 Max Turbo Pricing
The table below uses fal’s public per-second rates as of September 3, 2026, keeping the platform consistent. These are not SeeAPI prices; check the live SeeAPI model page before purchasing or scaling usage.
Model | 480P | 768P | 2K | Five seconds at 768P |
|---|---|---|---|---|
MiniMax H3 | $0.05/sec | $0.06/sec | $0.13/sec | $0.30 |
H3 Max | $0.05/sec | $0.08/sec | Not available | $0.40 |
H3 Max Turbo, launch rate | $0.00625/sec | $0.01/sec | Not available | $0.05 |
H3 Max Turbo, standard rate | $0.025/sec | $0.04/sec | Not available | $0.20 |
fal states that Turbo’s 75% launch discount ends September 7, 2026. Rates vary by platform and promotion. Compare cost per usable clip: a cheaper model that needs more retries may cost more in production.
Which MiniMax H3 Model Should You Use?
Choose MiniMax H3 for 2K delivery, advanced editing, multimodal reference packs, or voice and motion transfer.
Choose H3 Max when 768P is sufficient and you want fal’s balance of prompt adherence, aesthetics, and sub-real-time generation.
Choose H3 Max Turbo for the fastest drafts, batch exploration, live prototypes, or cost-sensitive iteration.
If uncertain, filter by required resolution first. Once 768P is acceptable, compare Max and Turbo on usable-output rate rather than latency alone.
MiniMax H3 vs H3 Max vs H3 Max Turbo: Verdict
The family is one open foundation with two speed-focused derivatives. H3 offers the widest control and highest official resolution. H3 Max favors faster 768P production; Turbo pushes latency and current pricing lower but has narrower documented workflows and less standalone evidence.
Start with the delivery requirement, then weigh reference complexity, waiting time, and cost per approved result. For a practical comparison, try MiniMax H3 Max on SeeAPI with one repeatable five-second prompt.
