NaiveAI releases open-weights model it says AI helped build

Summary

AI startup NaiveAI has released Naive-N0.5-Flash, an open-weights model for coding and AI R&D. The 309B-parameter MoE model activates 15.5B parameters, supports a native 1M-token context through a mix of sliding-window and sparse attention with no full-attention layers, and ships under the MIT license. The company says AI systems handled substantial parts of the architecture design, training, and inference optimization, while human researchers set direction and made key decisions. Performance figures such as 2,000 tokens per second are company claims, not independently verified.

Why it matters for our work

NaiveAI runs AI-centered R&D at production scale, with people setting direction and AI running experiments and implementation. It is a signal of how far the division of labor between people and AI can move in research and engineering work.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.