Nemotron 3.5 Lightning vs Qwen3.6: which 3B-active model to use?

Nemotron 3.5 Lightning is a 30B-A3B hybrid for one H100 with 1M context; Qwen3.6 is a 35B-A3B plus 27B dense pair with image input. Compare fit and results.

  • Updated

Verdict. Pick Qwen3.6 if you want the stronger coding and agent results, and Nemotron 3.5 Lightning if you want a 1M-token context, speculative decoding on NVIDIA hardware and the permissive OpenMDW license. Lightning is a 30B-A3B hybrid mixture of experts for one 80 GB H100 under the OpenMDW license. NVIDIA's own table shows Qwen3.6-35B-A3B well ahead on coding and agentic tests, and only six languages are listed for Lightning. Qwen3.6 is a 35B-A3B mixture of experts plus a 27B dense model under Apache 2.0, with a 256K context and image and video input. Both tables come from the developers.

Specifications

Specifications Nemotron 3.5 LightningQwen3.6
DeveloperNVIDIAAlibaba Qwen
ReleasedAugust 11, 2026April 16, 2026
Total parameters31.6B (Lightning 30B-A3B (NVFP4))36B (35B-A3B)
Active parameters3B3B
ArchitectureMixture of expertsMixture of experts
Context window1M tokens256K tokens
InputTextText, Image, Video
License OpenMDW-1.1 (Permissive) Apache 2.0 (Permissive)
Memory for the smallest variant (4-bit)22 GB (Lightning 30B-A3B (BF16))19 GB (27B)

Performance claims come from the developer and have not been independently verified.

Pick Nemotron 3.5 Lightning if

  • You run NVIDIA GPUs, especially a single 80 GB H100, and want fast agent and chat workloads.
  • You work with very long documents, since it has 256K context on one H100 and up to 1M on larger setups.
  • You want open recipes for fine-tuning or distilling a hybrid mixture of experts.
  • You want to serve with speculative decoding on Blackwell hardware.

Pick Qwen3.6 if

  • Coding and agent quality matters most, since NVIDIA's own table puts Qwen3.6 well ahead.
  • You need image or video input.
  • You want a 27B dense option for a 24 to 48 GB GPU or a high-memory Mac.

Nemotron 3.5 Lightning · Qwen3.6