New open-weight releases, retirements, and platform changes. Retirements are announced here at least 30 days before the retirement date — see Model lifecycle.
Catalog refresh and three inference regions
- Added
Kimi and MiniMax families, including image and video input on Kimi K3 and MiniMax M3.
- Added
Latest DeepSeek and GLM releases. DeepSeek V4.1 Flash accepts image input; GLM 5.3 Flash accepts image and video.
DeepSeek V4.1 FlashDeepSeek V4 Pro 0813DeepSeek V4 Flash 0731GLM 5.3GLM 5.3 FlashGLM 5.2
- Added
Qwen3.8 and Xiaomi MiMo V2.6.
Qwen3.8 27BQwen3.8 FlashQwen3.8 2.4T A95BQwen3.6 35B A3BMiMo-V2.6-ProMiMo-V2.6-Flash
- Removed
Earlier releases replaced by the models above, before any traffic was served: DeepSeek V3.2, DeepSeek V4 Flash (April release), GLM 5.1, Qwen3.5 397B and 122B, MiMo V2.5 Pro.
- Platform
Inference now runs in three regions: Johor (primary), Jakarta, and Dublin. See Regions.
- Added
Pre-launch catalog and documentation
- Platform
Published the pre-launch model catalog, API reference, and zero-data-retention documentation.
- Platform