The Open Weights
LatestModelsLeaderboardsCompanies
Subscribe
The Open Weights

The daily record of open-source AI. New model releases, leaderboards, and what's coming next — written for people who ship.

Refreshed every 12 hours

Discover

  • Latest releases
  • New today
  • Trending models

Browse

  • All models
  • Companies
  • Categories
  • Leaderboards

About

  • About
  • Editorial policy
  • Hardware estimates
  • RSS feed
  • llms.txt
  • Newsletter

© 2026 The Open Weights. An independent publication.

PrivacyTermsSMSAggregated by Claude · curated by humans.
LatestNVIDIA3.5-Lightning
NVIDIAText / LLM

NVIDIA's Nemotron 3.5 Lightning trims MoE for speed

A 30-billion-parameter mixture-of-experts model activates just 3 billion parameters per token, using a hybrid Mamba design to keep inference fast.

Aug 4, 2026
NotableOther
Nemotron-3.5-Lightning-30B-A3B

NVIDIA has published Nemotron-3.5-Lightning-30B-A3B, a new open-weights language model that leans on a mixture-of-experts (MoE) design to balance capability against inference cost. The model carries 30 billion total parameters but activates only about 3 billion per token, a ratio that lets it punch above the compute budget of a dense model its size. It is available now on Hugging Face.

The "Lightning" branding points to the model's focus on throughput. Alongside the sparse MoE routing, NVIDIA uses a hybrid Mamba architecture, blending state-space layers with attention to reduce the quadratic cost that traditional transformers pay on long sequences. The result is a model built for faster generation without discarding reasoning ability, which is listed among its core modalities.

Why it matters

Efficiency-first releases are becoming NVIDIA's signature in the open model space, and this one packages several of those ideas together:

  • A 30B/3B-active MoE that keeps per-token compute low
  • A hybrid Mamba backbone for cheaper long-context handling
  • NVFP4 quantization, tuned for NVIDIA's own hardware, to shrink the memory footprint
  • Multilingual coverage rather than English-only training

The NVFP4 format is the notable detail here. As a 4-bit floating-point scheme aligned with NVIDIA's latest accelerators, it signals that the company is optimizing not just for accuracy benchmarks but for how these models actually run on its chips. For teams weighing self-hosted reasoning models, that combination of sparse activation and aggressive quantization is the pitch.

NVIDIA has not published a context length for this variant, and the model ships under a custom license rather than a standard open-source one, so prospective users should review the terms before building on it.

Sources

  • nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4

    Hugging Face

    Visit
OlderMistral debuts Shieldstral, a 3B safety modelMistral AI · Vision-Language · 4 months agoNewerDeepgrove's Maple Preview bets on ternary-weight MoEDeepgrove · Reasoning · 4 months ago

Get the model

Hugging Face

Specs

Parameters30B · MoE
Active params3B active
Context window1.0M tokens
Size19.7 GB
PrecisionINT8
ArchitectureNemotronHForCausalLM
LicenseOTHER
Downloads907K
Likes280

Can you run it?

Runs on a laptop — about 12.4 GB at 4-bit.

  • INT8 (as published)

    24 GB GPU (RTX 3090 / 4090) · 32 GB Mac

    21.4 GB
  • 4-bit

    16 GB GPU (RTX 4080 / 5070 Ti) · 24 GB Mac

    12.4 GB
Your machine
GGUF builds
  • Mixture-of-experts: all 17.8B parameters must be resident even though only 3B are active per token — active count affects speed, not memory.
  • KV cache approximated from model size (no config metadata) at 8,192 tokens.

Based on Hugging Face weights metadata. Assumes an 8K context and ~1 GB runtime overhead; actual needs vary. How we estimate


Modalities

Text / LLMReasoning
2 versions — view changelog

The Weekly Weights

Every open release that mattered, one email a week.

0 comments

No comments yet. Be the first to weigh in.

More from NVIDIA

All NVIDIA releases →
NVIDIA/Text → Video

NVIDIA's Cosmos 3 Brings Open Interactive World Models

The new open release pushes beyond passive video generation toward simulated worlds that respond to input.

Nov 5, 2026
NVIDIA/Image → Video

NVIDIA Releases Cosmos 3 for World Simulation

The new open models push NVIDIA's world-foundation effort toward interactive, real-time video generation.

Nov 5, 2026
Nemotron Labs Dory 1.0 4B
NVIDIA/Text / LLMRuns on a laptop

NVIDIA releases compact Dory 1.0 4B text model

The new Nemotron Labs model packs a 4-billion-parameter language model trained on NVIDIA's Nemotron datasets.

Oct 16, 2026

More in Text / LLM

All Text / LLM →
Aion-1.0-Instruct-2.3B
Microsoft/Text / LLMRuns anywhere

Microsoft's Aion 1.0 Instruct is a compact 2.3B chat model

A small, multilingual instruction-tuned model aimed at chat and summarization lands on Hugging Face.

Nov 18, 2026
SenseNova-Loop-9B
SenseTime/Vision-LanguageRuns on a laptop

SenseTime Releases Looped 9B Multimodal Model

The company's new SenseNova-Loop-9B pairs vision-language capabilities with a recurrent looped-transformer design.

Nov 17, 2026
Hunyuan Hy4-flash
Tencent/Vision-LanguageMulti-GPU server

Tencent releases Hunyuan Hy4-flash multimodal MoE

The lightweight, mixture-of-experts model adds vision and text instruction-following to Tencent's open Hunyuan lineup.

Nov 17, 2026