NaiveAI/Naive-N0.5-Flash

NaiveAI publishes a 309B parameter mixture-of-experts model with 15.5B active weights, licensed under MIT for coding and research tasks. The architecture supports a native one-million token context window using hybrid sliding-window and sparse attention mechanisms. Inference code is included, with API access planned at low per-token rates. Independent verification of the claimed 2,000 tokens per second throughput remains pending.