Skip to main content

Nemotron 3.5 Lightning (30B A3B)

NVIDIAOpen WeightOpenMDW License v1.1 · Commercial OK

Description

NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts (MoE) model with ~3B active parameters, built for high-throughput specialized task execution in long-running / always-on agents. It ships with speculative decoding (MTP, DFlash, DSpark), NVFP4 and BF16 checkpoints, and is designed to pair with NVIDIA NeMo Switchyard for routing execution work alongside frontier planners. Released with open weights, data, and recipes under OpenMDW-1.1.

Release Date
2026-08-11
Parameters
30.0B
Context Length
262K
Modalities
text

Capability Radar

50
general
50
coding
40
reasoning
43
scienceest.
20
agents
0
multimodal

Science is estimated from LLM Stats science scores or reasoning when no dedicated science benchmarks are available.

Rankings

Domain#RankScoreSource
Agentic Capability87
38.0
LS

Benchmark Scores (LLM Stats)

(LLM Stats (zeroeval))

Agents

Tau3 Banking9.3%SR

Code

PinchBench85.4%SR

Instruction Following

IFBench71.9%SR

Language

MMLU-Pro81.9%SR

Reasoning

GPQANYU + Cohere + Anthropic (2023)75.4%SR
AA-LCR52.0%SR
SWE-Bench Verified51.6%SR
SWE-bench Multilingual39.3%SR
BrowseCompOpenAI (2025)37.0%SR
SciCode32.6%SR
Terminal-Bench 2.124.6%SR
Humanity's Last Exam11.7%SR

Science

OmniScience17.5%SR

AA Evaluation Indices

(Artificial Analysis)

No AA evaluation data available

LLM Stats Category Scores

(LLM Stats (zeroeval))
Language
80
Legal
80
Finance
80
Healthcare
80
Instruction Following
70
Long Context
50
Physics
50
Frontend Development
50
General
50
Biology
50
Chemistry
50
Code
50
Math
40
Reasoning
40
Search
40
Agents
40
Knowledge
20
Science
20
Tool Calling
20
Vision
10

Pricing

Input Price$0 / 1M tokens
Output Price$0 / 1M tokens
Blended Price (3:1)$0 / 1M tokens

Speed

No speed data available

Provider Price Ranking

Provider Price Ranking

2 providers

Cheapest: NVIDIAMost Expensive: DeepInfra
ProviderInputOutput
1NVIDIAPRIMARY
$0
$0
2DeepInfra
$0
$0

Compare pricing across different API providers for this model.

External Sources