NVIDIA Open-Sources Nemotron-Labs-Sigma Diffusion Models on Hugging Face
11/20/2026 — 11/21, 20:17·2 sources (all)·2 reports (all)
Key takeaways
- NVIDIA has released Nemotron-Labs-Sigma, a family of open-weight continuous diffusion language models on Hugging Face, offering 14 checkpoin…
Background & analysis
On November 20, 2026, NVIDIA announced Nemotron-Labs-Sigma, a family of continuous diffusion language models shipped as 14 open checkpoints spanning 3B and 8B sizes. According to a MarkTechPost report that day, the models denoise text inside a learned 16-D embedding space, can be distilled into few-step samplers, and decode up to 4.20x faster than autoregressive decoding.
A follow-up report from DEV Community · AI on November 21, 2026 added further details on the release. It described Nemotron-Labs-Sigma as an open-weight family of continuous diffusion language models, with its 14 checkpoints split across 3B and 8B sizes. The set covers base, instruct, and few-step distilled variants labeled NFE 2, 4, 8, and 16, plus one self-speculative variant per size. All weights were published on Hugging Face under the NVIDIA Nemotron Open Model License, a license that permits commercial use, so teams can download, fine-tune, and serve the models in their own environments. The report also stressed that, unlike token-by-token autoregressive or discrete diffusion approaches, Sigma denoises in the learned 16-D embedding space and generates in chunks of 64 tokens each.
As of the available coverage, the story stands at the point where the model weights and licensing terms have been made public, with no further updates reported yet.
AI-generated from 2 reports · updated 14 hours ago
Latest turnNVIDIA released Nemotron-Labs-Sigma on Hugging Face, an open-weight family of continuous diffusion language models shipping 14 checkpoints across 3B and 8B sizes: base, instruct, few-step distilled variants at NFE 2, 4, 8 and 16, plus a self-speculative variant per size. All weights fall under the NVIDIA Nemotron Open Model License, which permits commercial use, so teams can download, fine-tune and serve them in their own environments. Unlike autoregressive models or discrete diffusion, Sigma denoises token representations in a learned 16-dimensional embedding space and generates text in blocks of 64 tokens.
- Heat index
- 101
- All sources
- 2
- All reports
- 2
- First seen
- 2 days ago
Reports on this story headlines open the original
NVIDIA released Nemotron-Labs-Sigma on Hugging Face, an open-weight family of continuous diffusion language models shipping 14 checkpoints across 3B and 8B sizes: base, instruct, few-step distilled variants at NFE 2, 4, 8 and 16, plus a self-speculative variant per size. All weights fall under the NVIDIA Nemotron Open Model License, which permits commercial use, so teams can download, fine-tune and serve them in their own environments. Unlike autoregressive models or discrete diffusion, Sigma denoises token representations in a learned 16-dimensional embedding space and generates text in blocks of 64 tokens.
DEV Community · AIAI score 55
NVIDIA has released Nemotron-Labs-Sigma, a family of continuous diffusion language models spanning 14 open checkpoints at 3B and 8B. The models denoise text in a learned 16-D embedding space, distill to few-step samplers, and decode up to 4.20x faster than autoregressive decoding.
MarkTechPostAI score 62
Other stories people are talking about
- Heat index 25Hugging Face releases harness adapters1 sources
- Heat index 1143RisingAnthropic Reaches $1.6 Trillion Market Value in First Trading Week After IPO21 sources
- Heat index 1011Anthropic Mathematician Says Claude Formalized the Generalized Riemann Hypothesis21 sources
- Heat index 622Anthropic IPO total raise reaches $93.5 billion after greenshoe exercise20 sources
- Heat index 596RisingThinking Machines Lab Open-Sources Inkling 2 Mini with 32GB Model Card Recommendation10 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

