Inkling 2 Mini vs WeMM-Video-2B
Compare documented capabilities and access. A dash means a comparable figure is unavailable.
| Specification | ||
|---|---|---|
| Task | Multimodal reasoning and coding | Audio-visual understanding |
| Published capabilities | Reuses its middle layers three times per token. Thinking Machines reports 48.2% on DeepSWE v1.1. | Answers questions in Chinese or English about video frames and their soundtrack. Supports direct responses and a thinking mode. |
| Model size | 26.4B total / 3.4B active parameters | 2B dense language backbone |
| Access | Downloadable weights and Tinker inference and customization | Public BF16 checkpoint and quantized releases |
| Result | — | Text responses, including descriptions and speech transcripts |
| Operating limits | Native mixed-precision checkpoint occupies approximately 20 GB. | — |
| Sources | Model weights · Additional documentation | Model weights · Additional documentation |
| Inputs | Text, Images | Text, Images, Audio, Video |
| Weights | Open weights · Apache-2.0 | Open weights · Apache-2.0 |
| Released | Nov 20, 2026 | Nov 20, 2026 |