NaiveAI Ships 309B MoE Model with Hybrid Sparse Attention

Beijing startup NaiveAI has released Naive-N0.5-Flash, an MIT-licensed 309B mixture-of-experts model featuring 15.5B active parameters and 1M-token context. The model, valued at $1.4B, was largely developed by AI systems with human oversight and claims high inference speeds. The news also briefly mentions MiniMax shipping its M3.1-Flash-Preview.