MiniMax H3: Multimodal Video Model Release
MiniMax released MiniMax H3 on July 31, 2026. This open-weights model supports text input and output while processing multimodal creative intent.
Release Details and Modalities
MiniMax released MiniMax H3 on July 31, 2026. The source identifies this event as a release of a new-generation, general-purpose multimodal video model. The announcement date is distinct from the current publication date of this article.
The model card indicates that MiniMax H3 has open weights. According to the official changelog, the model understands creative intent across multimodal context, specifically listing text, image, video, and audio. The documented input modality is text, and the documented output modality is text.
Documented Capabilities vs. Unknowns
The source establishes that the model delivers more natural and coherent generation and expression. However, the provided facts do not specify pricing, context window size, or maximum output tokens. No performance benchmarks, latency metrics, or comparative rankings against previous models are included in the source material.
While the model is labeled as multimodal regarding its understanding of creative intent, the explicit input and output modalities listed in the data are limited to text. The source does not confirm whether direct image, video, or audio inputs are accepted via API endpoints, nor does it detail specific architectural components or training methodologies.
Evaluation Procedure for Engineers
Engineers can evaluate MiniMax H3 by designing tests that validate its stated ability to understand creative intent across multimodal contexts. Since the source confirms open weights, users may obtain the model artifacts from the referenced repository.
- Test the model’s response coherence by providing text prompts that reference hypothetical visual or audio scenarios.
- Measure the consistency of the generated text when describing complex creative intents involving multiple media types.
- Compare the output quality against baseline expectations for natural language generation without assuming specific speed improvements.
- Validate that the model handles edge cases in multimodal context interpretation as described in the release notes.
Primary sources
Frequently asked questions
What are the confirmed input and output modalities for MiniMax H3?
The source explicitly lists text as both the input and output modality. While the model is described as understanding creative intent across text, image, video, and audio contexts, the structured data only confirms text-based interaction.
Is MiniMax H3 available with open weights?
Yes, the source facts indicate that openWeights is true for MiniMax H3. Evidence includes an official model card observed on Hugging Face under the MiniMaxAI organization.