On 31 July 2026, MiniMax announced MiniMax H3. MiniMax unified text, images, video and audio as context and generated video with synchronised native stereo audio in one system.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 31 July 2026 |
| Availability or weight release | 31 July 2026 by product and API; H3 Base checkpoints published openly |
| Release type | open-weight omni-modal audio-video generation system |
| Access | H3 Base weights under the MiniMax H3 Community Licence; Context-IR and 2K regeneration remain hosted |
| Architecture | a 33B dense single-stream omni-transformer with separate visual and audio VAEs and a Qwen3-VL-32B encoder |
| Maximum stated context | multimodal reference inputs with up to 15 seconds of source audio or video per clip |
What changed
H3 is a major open generative-model release because it exposes the base audio-video transformer rather than only a hosted demo. The complete product still has three parts: hosted Context-IR, open H3 Base at 768p and hosted in-context regeneration for 2K. That boundary must be preserved in any claim of local reproducibility. The model can produce four-to-fifteen-second clips at 24 frames per second with 32kHz stereo audio.
The practical comparison is therefore not simply whether MiniMax H3 has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| Output duration | 4–15 seconds | Published capability envelope, not a quality score |
| Output format | up to 2K, 24 FPS, 32kHz stereo | 2K requires the hosted regeneration stage |
| Training throughput | nearly 30% improvement | MiniMax architecture-and-systems claim |
These are release-time results, not independently reproduced guarantees. MiniMax did not publish a conventional numerical quality benchmark table in the initial card, and the full official 2K pipeline is not wholly open. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
MiniMax H3 is described as a 33B dense single-stream omni-transformer with separate visual and audio VAEs and a Qwen3-VL-32B encoder with multimodal reference inputs with up to 15 seconds of source audio or video per clip of stated context. Its access position at verification time is H3 Base weights under the MiniMax H3 Community Licence; Context-IR and 2K regeneration remain hosted. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For MiniMax H3, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
H3 belongs in the archive despite its specialist modality because it is a substantial open-weight system. Teams should distinguish local 768p Base results from hosted 2K outputs and review the custom licence before deployment.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: MiniMax official release artwork. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
