Kimi Linear
by Moonshot AI
· Kimi family · released Oct 30, 2025
Open weights
Active
Kimi Linear is a 48B-total/3B-active hybrid linear attention model using Kimi Delta Attention, supporting up to 1M-token contexts with higher throughput and better long-context recall than full-attention baselines.
No pricing recorded for this model yet.
Specifications
| Model type | Text LLM |
|---|---|
| Architecture | Hybrid |
| Parameters | 48 (3 active) |
| Context window | 1M tokens |
| Max output | Not disclosed tokens |
| Knowledge cutoff | Not disclosed |
| License | Modified MIT |
| Modalities | Text input Text output |
| Tags | Long context Small LM |
| End of life | Not disclosed |
| Last verified | Not disclosed |