Journalism begins where hype ends

,,

The danger of AI is not that it will become conscious and hate us, but that it will become competent and ignore us."

—Eliezer Yudkowsky

NaiveAI Launches Naive-N0.5 Flash, a 309B Open-Weight AI Model With 1M Context

Beijing-based AI startup NaiveAI has launched Naive-N0.5 Flash, a 309B open-weight Mixture-of-Experts model with 15.5B active parameters and a native 1 million-token context window, targeting coding and AI research.
NaiveAI Naive-N0.5 Flash 309B open-weight AI model logo
September 28, 2026 12:51 PM IST | Written by Vaibhav Jha

Beijing based AI startup NaiveAI launched its flagship Naive-N0.5 Flash, an open-weight 309B Mixture-of-Experts (MoE) model with 15.5B active parameters with 1 million native context.

Naive-N0.5 Flash, an open-weight AI model is built on Xiaomi MiMo-V2.5 base model and then further trained on 3.25 trillion tokens.

The creators of Naive-N0.5 Flash claim the model was built under the ecosystem of AI centered R&D where AI models write code, run experiments, monitor progress, analyze results, and iterate. While human researchers set direction and make critical decisions.

“AI explored and designed its hybrid attention architecture while optimizing its training, inference, and deployment systems. Human researchers provided guidance and made key decisions. Naive-N0.5-Flash is also trained for AI R&D, allowing it to participate directly in the R&D process and opening a path toward recursive self-improvement (RSI),” read a statement from NaiveAI Lab.

 

According to NaiveAI Lab, the Naive-N0.5-Flash combines Sliding-Window Attention (SWA) and lightweight DeepSeek Sparse Attention (DSA) with GQA4 at a predominantly 5:1 SWA–DSA layout.

This allows the AI model to read extremely long documents or codebases without the high compute cost as it handles a million tokens of context without any traditional “full attention” layers by mixing short sliding-window attention with a sparse selection method.

NaiveAI claimed that the flagship model is built for coding and AI R&D and announced that API access will also be provided, with pricing set at $0.10 / $0.40 / $0.01 per million tokens for input, output, and cache reads, respectively.

Also Read: Meta Launches Muse Glimmer 30B Open-Weight Local AI Model

Author

  • Vaibhav Jha, editor and co-founder at AI FrontPage

    Vaibhav Jha is an Editor and Co-founder of AI FrontPage. In his decade long career in journalism, Vaibhav has reported for publications including The Indian Express, Hindustan Times, and The New York Times, covering the intersection of technology, policy, and society. Outside work, he’s usually trying to persuade people to watch Anurag Kashyap films.

    LinkedIn

Related Posts

LATEST NEWS