Skip to main content
Language model

Nemotron 3.5 Lightning 30B A3B

NVIDIA

Released
August 11, 2026
Data date
September 26, 2026

Nemotron 3.5 Lightning 30B A3B is the BF16 checkpoint in the Lightning family. It activates 3 billion of 30 billion parameters and uses the hybrid Mamba2, MoE, and attention architecture with multi-token prediction. The model returns text.

NVIDIA advertises up to 1M context tokens, but lists a memory-bound 256K limit on a single H100. The BF16 configuration sets max_position_embeddings to 262,144. According to the model card, the larger window was validated on multiple GPUs.

Specifications and access

SpecificationValue and source
Weights
Open weightsSource
Access
APISource
Parameters
30 billionSource
Active parameters
3 billionSource
License
NVIDIA Open Model LicenseSource
Inputs
TextSource
Outputs
TextSource
Reasoning
YesSource
Tool calling
YesSource
Advertised context
Up to 1M tokensSource
Packaged default configuration
262,144 tokensSource