AnalysisAI ModelsJuly 25, 2026
Kimi Linear 48B MoE model spotted with 1M context
A Reddit user discovered a new MoE model named 'Kimi Linear' with 48B total parameters (3B active) and 1M context. It runs fast compared to Qwen 3.6 35B but tends to produce minimal output.
1 source
More stories today
- No sign of Anthropic joining OpenAI coalition
- LangChain argues companies must own their AI systems for lasting advantage
- Robot company ships distilled world model to millions
- Chollet predicts end of big model launches within 2 years
- Rauch shares agent CLI research workflow with _open.yml