Inkling 2 Mini vs Aion 1.0 Instruct 2.3B
Compare documented capabilities and access. A dash means a comparable figure is unavailable.
| Specification | ||
|---|---|---|
| Task | Multimodal reasoning and coding | On-device language generation |
| Published capabilities | Reuses its middle layers three times per token. Thinking Machines reports 48.2% on DeepSWE v1.1. | Windows-oriented instruction model for local text generation, including document and email tasks. |
| Model size | 26.4B total / 3.4B active parameters | 2.3B parameters |
| Access | Downloadable weights and Tinker inference and customization | Downloadable checkpoint |
| Operating limits | Native mixed-precision checkpoint occupies approximately 20 GB. | 32K-token context window |
| Sources | Model weights · Additional documentation | Model weights |
| Inputs | Text, Images | Text |
| Weights | Open weights · Apache-2.0 | Open weights · MIT |
| Released | Nov 20, 2026 | Nov 18, 2026 |