HUMAIN M3
HUMAIN M3 adapts MiniMax-M3 for Arabic. HUMAIN's announcement describes continued training on more than one trillion Arabic-native tokens. The current release is a preview with an API and playground; public weights are still listed as coming soon.
Preview
The two preview modes expose different behavior
Limited preview adds HUMAIN's Saudi alignment filter and disables thinking and streaming. Approved research access enables thinking and streaming with lower stated latency. A model feature in the announcement is therefore not necessarily enabled for every account.HUMAIN Node current M3 specifications and access
Preview interactions become evaluation data
The model terms state that prompts and responses are recorded and can be reviewed for model development. Consent is required, and the terms prohibit sensitive or confidential inputs. This research environment has different conditions from a private deployment.HUMAIN M3 model terms
A planned weight release is separate from an available download
The September announcement planned an open-weight release. Node still lists it as coming soon. Preview access therefore does not establish download access.HUMAIN's M3 announcement and MiniMax attributionHUMAIN Node current M3 specifications and access
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | API, Provider product |
| Model identifiers | humain-m3 |
| Model size | 428 billion total parameters; 23 billion active per token in a mixture-of-experts design.HUMAIN Node current M3 specifications and access |
| Inputs | Text, images and video. The published multimodal claim is about understanding, not an image-generation endpoint.HUMAIN Node current M3 specifications and access |
| Preview purpose | Evaluation and non-production prototyping. Approval and account limits apply.HUMAIN Node operator and preview conditions |
| Public weights | Not provided by the preview. The model terms do not commit HUMAIN to a later weight release.HUMAIN M3 model terms |
Related models
The Arabic and English model introduced with HUMAIN Chat, building on the earlier ALLaM research program at NCAI and SDAIA.
Launch date not confirmedProvider product- ALLaM 7B Instruct Preview
HUMAIN · ALLaM
NCAI's downloadable Arabic and English 7B model, now published through HUMAIN, with a 4,096-token context and documented revision caveats.
Launch date not confirmedDownloadable