Nemotron 3.5 Lightning

Name: Nemotron 3.5 Lightning
Price: 0.049 USD
Author: NVIDIA

NVIDIA · Nemotron 3.5 · released 2026-08-11 · data checked 2026-11-01 Needs verification

Nemotron 3.5 Lightning is a 31.6B-parameter Mixture-of-Experts model (3.6B active per token) from NVIDIA (United States), released 2026-08-11 under the OpenMDW-1.1 license. It scores 63.6 on the Small LM Index, #14 of 34 medium models. It needs about 22.4 GB of memory at 4-bit with an 8K-token context and costs $0.049 per million tokens through the cheapest API provider.

Hybrid Mamba-Transformer MoE tuned for throughput (~670 tok/s measured by Artificial Analysis).

Small LM Index

63.6

#14 of 34 scored medium models

Likely 55.3–70.9 · from 3 of 6 core benchmarks

Parameters

31.6B

3.6B active · MoE

Price per 1M tokens

$0.049

$0.025 in / $0.12 out · blended 3:1

Context window

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 3 reported by NVIDIA. How we label sources.

MMLU-Pro Lab-reported Knowledge81.9

Class best: Qwen3.6 27B 86.2

GPQA Diamond Lab-reported Knowledge75.4

Class best: Nanbeige4.3 33B 89.5

SWE-bench Verified Lab-reported Coding51.6

Class best: Qwen3.6 27B 77.2

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 13

AIME≈ 88

LiveCodeBench≈ 73

Specifications

Lab: NVIDIA (US · United States)
Architecture: Mixture-of-Experts
Total parameters: 31.6B
Active parameters: 3.6B
Size class: Medium
Context window: 1M tokens
Input modalities: text
Reasoning: Reasoning (thinking)
License: OpenMDW-1.1
Openness: Open weights
Memory to run: ~22.4 GB at 4-bit, 8K context (more sizes)
API providers: 5 on OpenRouter
API price: $0.025 in / $0.12 out per 1M (OpenRouter, 2026-11-03)

Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

Nemotron 3.5 Lightning vs GLM-4.7-Flash (31B, index 66.1)
Nemotron 3.5 Lightning vs Nemotron 3 Nano 30B-A3B (31.6B, index 62.6)
Nemotron 3.5 Lightning vs EXAONE 4.0 32B (32B, index 62.0)
Nemotron 3.5 Lightning vs Sarvam 30B (32.2B, index 61.6)
Nemotron 3.5 Lightning vs OLMo 3.1 32B Think (32.2B, index 57.0)

Ranked in

#8 in Best cheap LLM APIs (open models)
#8 in Best American open models under 100B

Sources

Something wrong here? Report it.

Run it locally

No quantized build from a well-known publisher yet; check the model card for instructions.

Memory needed: weights, context cache and runtime buffers (context cache estimated: attention layout not published).

Quantization	4K context	32K context	256K context
16-bit F16	65 GB	73 GB	132 GB
8-bit Q8_0	36 GB	43 GB	103 GB
4-bit Q4_K_M	21 GB	29 GB	88 GB

Check it against your GPU, Mac or phone →