Qwen3.6 35B A3B (NovitaAI)

Qwen3.6 35B A3B (NovitaAI) is a multimodal Mixture-of-Experts model built around a sparse activation pattern of 35 billion total parameters with roughly 3 billion parameters activated per token, pairing broad capability coverage with the compute savings typical of MoE routing. NVIDIA's NGC catalog lists the model under the Qwen organization and offers it as a downloadable open-weights distribution, including signed NVIDIA NIM container variants such as the nim-spark-nvfp4-fast-mtp build, which makes the weights accessible to teams that prefer self-hosting over managed endpoints.

The NovitaAI-served entry on third-party pricing trackers documents a 262,144-token context window paired with a 64,000-token maximum output, placing the model firmly in the long-context tier suitable for multi-document analysis, repository-scale code tasks, and extended dialogue. Its multimodal MoE design suggests a balanced fit for workflows that mix several input types with the cost efficiency of sparse activation, while the open distribution channels give developers flexibility to run the weights through NIM-optimized containers or through gateway providers like Novita depending on operational needs.