model releaseNVIDIA

NVIDIA Releases Nemotron 3.5 Lightning: 30B MoE Model with 1M Token Context and 3B Active Parameters

TL;DR

NVIDIA released the full-precision BF16 reference weights for Nemotron 3.5 Lightning, a 30B-parameter Mixture-of-Experts model with only 3B active parameters and support for up to 1 million tokens of context. The model uses a hybrid Mamba-2, MoE, and Attention architecture and is licensed under OpenMDW-1.1 for commercial use.

2 min read
0

Nemotron 3.5 Lightning at a glance

Context window1049K tokens
Input$0.07/1M tokens
Output$0.2/1M tokens

How Nemotron 3.5 Lightning ranks now

BenchmarkScoreRankSource
AA Intelligence Index5.5 index45 of 49Artificial Analysis, Nov 2026 ↗
GPQA75.6%127 of 163Hugging Face model card ↗
SWE-bench Verified52.8%77 of 80Vals AI ↗
Speed (tok/s)270 tok/s5 of 101Artificial Analysis, Nov 2026 ↗
MMLU-Pro81.9%42 of 74Hugging Face model card ↗

Rank is against every model in our database with a sourced score on that benchmark, each counted once with its best score.

All Nemotron 3.5 Lightning specs and scores →

NVIDIA Ships Nemotron 3.5 Lightning as Open-Weight Reference Model

NVIDIA has released NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16, a 30-billion-parameter Mixture-of-Experts (MoE) language model with only 3 billion active parameters per forward pass. The model uses a hybrid architecture interleaving Mamba-2 layers, MoE layers, and select Attention layers, and supports context windows up to 1 million tokens.

This BF16 release is the full-precision reference checkpoint, intended primarily as a starting point for customization — supervised fine-tuning (SFT), reinforcement learning, distillation, domain adaptation, and building quantized variants — rather than for direct production inference. NVIDIA is separately shipping an NVFP4-quantized version for latency- and throughput-optimized deployment.

Specifications

  • Total parameters: 30B (3B active)
  • Architecture: Hybrid Mamba-2 + MoE + Attention, with Multi-Token Prediction (MTP) layers
  • Context length: Up to 1M tokens (validated at 256K on a single H100 80GB; full 1M requires 8x H100 TP8+EP or Blackwell GB200/B200)
  • Precision: BF16
  • Hardware support: NVIDIA Blackwell (GB200, B200), Hopper (H100, H200), Ampere (A100)
  • Languages: English, Spanish, French, German, Italian, Japanese, plus coding languages
  • Reasoning mode: Configurable on/off via chat template (enable_thinking=True/False)
  • License: OpenMDW License Agreement v1.1, commercial use permitted
  • Pre-training data cutoff: September 2025; post-training data cutoff: May 2026
  • Pre-training scale: Over 20 trillion tokens, using an NVFP4 pre-training recipe

Benchmark Results

According to NVIDIA, evaluated under its internal NeMo Gym / NeMo Evaluator SDK harness, Nemotron 3.5 Lightning scored:

  • MMLU Pro: 81.94
  • GPQA Diamond (no tools): 75.44
  • HLE (text-only, no tools): 11.72
  • SWE-bench Verified: 51.56
  • SWE-bench Multilingual: 39.33
  • Terminal-Bench 2.1: 24.58
  • IFBench (loose): 71.88
  • AA-LCR (long context): 52.00

NVIDIA's own comparison table places these scores below its larger Nemotron 3 Super model and below Alibaba's Qwen 3.6 35B-A3B on most reasoning and coding benchmarks, though Nemotron 3.5 Lightning outperforms GPT-OSS 20B on several metrics including MMLU Pro, GPQA Diamond, and IFBench. NVIDIA states these numbers may differ from vendors' self-reported figures due to harness differences.

Deployment

NVIDIA provides a vLLM-based quick-start command for single-H100 deployment, using vLLM version v0.27.1 with prefix caching, async scheduling, and a Mamba-specific backend (flashinfer). The company recommends a sampling temperature of 1.0 and top-p of 0.95.

Pricing for API access was not disclosed, as this is an open-weight release rather than a hosted inference product.

What This Means

Nemotron 3.5 Lightning targets developers who want to customize and fine-tune a mid-size model rather than deploy it as-is. The 3B active-parameter footprint on a 30B total model gives inference costs closer to a much smaller dense model while retaining larger model capacity — the core appeal of MoE designs. Benchmark scores position it below flagship competitors like Qwen 3.6 35B-A3B on coding and agentic tasks, but NVIDIA appears to be optimizing for hardware efficiency and post-training flexibility rather than raw leaderboard rank. The parallel NVFP4 release signals NVIDIA is explicitly separating "customization" and "deployment" checkpoints, a packaging approach likely to become more common as MoE architectures proliferate.

Related Articles

product update

NVIDIA launches Nemotron for Science with Nemotron 3.5 Super VL and a chemistry dataset of nearly 50,000 questions

NVIDIA introduced Nemotron for Science on November 16, 2026. It pairs the open Nemotron 3.5 Super VL model with scientific evaluations, post-training recipes and LitmusBench v0.2, a chemistry dataset of nearly 50,000 questions.

research

NVIDIA recipe: RL post-training lifts Nemotron 3.5 Super VL accuracy a further 17% on drug-metabolism task

NVIDIA has published a recipe for post-training Nemotron 3.5 Super VL on scientific tasks using NeMo Gym and NeMo RL. In a drug-metabolism example, supervised fine-tuning improved accuracy by 12% and RL post-training lifted it a further 17% at a 16,384-token output budget.

model release

Endless Frontier open-sources Fx-Bio-1031, a 125B MoE bioinformatics model with 512K context

Endless Frontier (无界前沿) has open-sourced Fx-Bio-1031, a bioinformatics model with 125B total parameters and 6B active per token, with a 512K context window. The company claims 70.59% Pass@5 on the Difficult subset of Anthropic's BioMysteryBench, against 47.06% for GPT-6-Astra and 52.94% for Claude Opus-5.

research

NVIDIA releases two synthetic Japanese datasets, Nemotron-Personas-Finance-Japan and Nemotron-PII-Japan

At NVIDIA Agentic AI Day Japan 2026 in Tokyo on November 10, NVIDIA presented two synthetic datasets for Japanese AI development. Nemotron-Personas-Finance-Japan targets financial AI and user simulation, and Nemotron-PII-Japan covers PII detection and masking across 8 domains. Both are published on Hugging Face.

Comments

Loading...