← All models

NVIDIA: Nemotron 3.5 Lightning

NVIDIA

budget

Capabilities

ToolsStructured OutputExtended ThinkingLong Context

Strengths

  • Extremely low inference cost and energy footprint
  • Open weights enable self-hosting and full data privacy
  • Fast token generation for agentic workloads
  • Extended context window (262K tokens) for long documents
  • Structured output and tool-use support

Weaknesses

  • Limited coding performance relative to peers
  • Lower reasoning and analysis capability than larger models
  • Speed score hindered by benchmark methodology, not practical latency
  • Not suitable for complex multi-step reasoning tasks

Pricing

Input / 1M tokens

$0.07

Output / 1M tokens

$0.19999999999999998

Context window

262.144k

Transparency

Open weights

10.0 / 10

Open training data

0.0 / 10

Open methodology

7.5 / 10

Licence openness

8.0 / 10

Provider disclosure

7.0 / 10

FMTI company score

N/A

Composite:6.5 / 10

Open-weight MoE model. NVIDIA publishes model architecture and weights publicly. Training methodology partially disclosed. Licence permits commercial use.

Sustainability

Inference energy

8.0 / 10

Training footprint

N/A

Provider infrastructure

6.5 / 10

Composite:7.2 / 10

Small active parameter count (3B) yields low per-inference energy. MoE architecture enables efficient inference. Training footprint unknown. NVIDIA operates own data centres with moderate efficiency standards.

Task Fitness

question answering
44%
code
40%
mathematics
75%
business communication
71%
vision
0%
analysis
51%
extraction
71%
long-form generation
62%
research
61%
reasoning
33%
summarisation
72%
translation
65%
conversation
71%