Naive-N0.5-Flash

People can run it for coding tasks and AI research and development.

Availability

Model weights and inference code are released.

Pricing model

API pricing is stated as $0.10 / $0.40 / $0.01 per million input, output, and cache-read tokens; API access will also be provided.

Deployment

Requires FP8-capable NVIDIA GPUs; weights occupy approximately 315 GB.

Privacy

Not established

Licence

MIT License

• Naive-N0.5-Flash has 309B total parameters and 15.5B active parameters. ^{[1] [2]}

  Naive-N0.5-Flash is an open-weight **309B MoE model with 15.5B active parameters**, built for **coding and AI R&D**.

• The model supports a native 1M-token context window. ^{[1] [2]}

  It supports a **native 1M-token context window** through a hybrid of Sliding-Window Attention (SWA) and lightweight DeepSeek Sparse Attention (DSA), with no full-attention layers.

• The model uses hybrid Sliding-Window Attention and DeepSeek Sparse Attention, with no full-attention layers. ^{[1] [2]}

  It supports a **native 1M-token context window** through a hybrid of Sliding-Window Attention (SWA) and lightweight DeepSeek Sparse Attention (DSA), with no full-attention layers.

• Model weights and inference code are released under the MIT License. ^{[1] [2]}

  - **Open weights and API.** Model weights and inference code are released under the MIT license.

• The model requires FP8-capable NVIDIA GPUs for the documented Transformers deployment. ^{[1] [2]}
• The stated API rates are $0.10 per million input tokens, $0.40 per million output tokens, and $0.01 per million cache-read tokens. ^{[1] [2]}

  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.

• New models and tools focus on typed decisions and long-context tasks Weekly brief 2026-W39
• Xiaomi MiMo introduces MiMo-V2.6-Flash-MOPD Daily brief 2026-09-27