NaiveAI open-sources Naive-N0.5-Flash: a 309B MoE built by AI itself, up to 2,000 tok/s inference
Xianbao_QIAN · x · 2026-09-28
NaiveAI has released and open-sourced (MIT license) Naive-N0.5-Flash, pitched as "building frontier AI with AI":
- Architecture: 309B MoE with 15.5B active parameters; native 1M context with no full-attention layers, combining SWA with lightweight DeepSeek Sparse Attention (DSA).
- Focus: coding and AI R&D; the model is trained to participate in the R&D process itself, opening a path toward recursive self-improvement (RSI).
- AI-driven R&D: AI explored the hybrid attention design and optimized training, inference, and deployment, while humans set direction and made key decisions. The supporting infrastructure serves 10 million sandboxes weekly with 100,000 concurrent at peak.
- Inference: AI-optimized runtime reaching up to 2,000 tokens/s in Ultrafast mode.
Weights are live on Hugging Face and GitHub, with an accompanying tech blog.
Related event: NaiveAI Open-Sources 309B MoE Model Built Largely by AI Itself(3 posts)→
More from Models
- Judgement small models become essential: four companies ship within eight weeks — FinanceYF5 · 2026-11-12
- Addy Osmani's guide to picking a Claude model: start with Opus at medium effort — bibryam · 2026-11-12
- Google expands academic research access program with free Gemini 4 Carbon for scientists — pushmeet · 2026-11-12
- SoftBank found 15,000 vulns with OpenAI's tool, then GPT-6 Astra surfaced 35,000 more — IgorCarron · 2026-11-12
- Kakao ships on-device small model Raon-1 nano, proven in PUBG and PlayinZOI — Kangwook_Lee · 2026-11-12
- Meta releases MetaEncoder: one bi-encoder for retrieval and decision-making via natural language — pmttyji · 2026-11-12