by NVIDIA

Nemotron-Labs-Sigma review, pricing and limits

NVIDIA's research family for diffusion text generation, in 8B and 3B sizes.

Best for
Research on parallel text generation
Not for
Long documents and retrieval pipelines
  • Available
  • Open weights

Should you use Nemotron-Labs-Sigma?

The 8B base model scores 65.12 on HumanEval and 88.57 on GSM8K in NVIDIA's own paper, and NVIDIA measures 120 to 140 tokens per second on one H100. It suits researchers studying few-step generation and distillation, not production assistants, since the base model skips instruction tuning.

Best for

  • Research on parallel text generation The card names sampling, few-step distillation, trajectory steering and fine-tuning as its intended uses.
  • Few-step distillation experiments NVIDIA ships distilled variants at fixed step counts and says quality degrades gracefully at low step counts.

Not for

  • Long documents and retrieval pipelines Prompt and output together are capped at 4,096 tokens.
  • Production use behind a hosted API Every deployment is one you run yourself, because NVIDIA operates no endpoint for it.

Can you use Nemotron-Labs-Sigma?

  • Free plan: not published.
  • Open weights: released (NVIDIA Nemotron Open Model License).

Nemotron-Labs-Sigma pricing

Pricing model
open-weights (self-hosted; no per-token API price published)
Free tier
Weights are free to download under the NVIDIA Nemotron Open Model License; you pay for your own GPU.
TierRate
Self-hosted (open weights)Free download; pay only for your own GPU

NVIDIA hosts no service for these models, so there is no per-token price. The weights are free to download, and the cost is the CUDA GPU you run them on.

Self-hosting requirements

Self-hosted
Yes
Recommended GPU
NVIDIA Ampere, Hopper, Lovelace or Blackwell GPUs with bfloat16; the card's speed figure came from one H100. Linux only, and CPU inference is not supported.

What Nemotron-Labs-Sigma can do

Nemotron-Labs-Sigma is NVIDIA's family of open continuous diffusion language models, released on 12 November 2026 in 8B and 3B sizes. Instead of writing one token at a time, it denoises whole blocks of text in parallel. The weights are on Hugging Face under the NVIDIA Nemotron Open Model License, and the context window is 4,096 tokens.

What goes in and what comes out

Reads

  • text

Writes

  • text

Capabilities

  • Audio inputNot supported
  • Audio outputNot supported
  • Video inputNot supported
  • VisionNot supported

Key features

  • Continuous diffusion in a 16-D embedding space

    Text is denoised from Gaussian noise, with the denoiser and token embeddings learned jointly.

  • Block-parallel decoding

    Blocks of 64 tokens by default, with a configurable number of denoising steps per block.

  • Few-step distillation

    Distilled variants at 2, 4, 8 and 16 steps; the paper reports about 8x on math and 4x on coding.

  • Two sizes, base and instruct

    8B and 3B models, with instruct variants for chat.

Show all 6 features
  • Self-speculative variants

    Distill-spec models accept 33% more draft tokens per round than Nemotron-Labs-Diffusion-8B on GSM8K, per the paper.

  • Plain PyTorch inference

    Loads through Transformers with trust_remote_code, with no custom kernels or training framework needed.

Known limitations

  • The 4,096-token window covers prompt and output together, which rules out long documents.
  • English is the only language the card lists as supported.
  • A CUDA GPU is required: the card says CPU inference gives unusable output.
  • The base model is not an instruction-following assistant.

Supported languages

  • English

Nemotron-Labs-Sigma benchmarks

Standout result

65.12

HumanEval pass@1 for the 8B base model (DDPM, 256 steps per block, 10-seed mean)

Source: arXiv 2610.02665, Table 2, read 17 November 2026

Benchmark scores

  • HumanEval vendor-reported · 17 Nov 202665.1%Small programs that must pass hidden tests, % passing.

A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.

Strengths and limits

Strengths

  • The paper's 8B base scores 88.57 on GSM8K in diffusion mode, above the 87.26 it reports for Nemotron-Labs-Diffusion-8B (vendor-reported).
  • Few-step distilled variants are published, so speed tests do not require training your own.
  • Open weights under a licence the card describes as ready for commercial or non-commercial use.
  • The paper reports 65.12 on HumanEval for the 8B base model in diffusion mode, from NVIDIA's own authors (vendor-reported, read 17 November 2026).
Show 2 more
  • NVIDIA ships distilled 2, 4, 8 and 16-step variants, built to cut denoising steps with little accuracy loss.
  • The weights, sampler code and paper are public, so researchers can inspect the whole pipeline.

Limits

  • Prompt plus output share a 4,096-token window, so retrieval and long-document work is out of reach.
  • The only scores found are NVIDIA's own paper figures, and no independent board listed the model on 17 November 2026.

Nemotron-Labs-Sigma alternatives

ModelLabContextIn / out per 1MReads and writes
Nemotron-Labs-Sigma This pageNVIDIA4K--text → text
DiffusionGemma 26B-A4BPick Nemotron-Labs-Sigma if you want continuous diffusion at 8B or 3B with distilled few-step variants; pick DiffusionGemma 26B-A4B if you need a much longer window from an open diffusion model.Google256K--text, image, video → text
Mercury 3Pick Mercury 3 if you want a hosted diffusion API with a 1M-token window; pick Nemotron-Labs-Sigma if you want weights you can run and study.Inception1M$0.10 / $0.35text → text
Ministral 3 8BPick Ministral 3 8B if you need a general assistant with a long window; pick Nemotron-Labs-Sigma if you are studying parallel decoding.Mistral AI256K$0.15 / $0.15text, image → text

Nemotron-Labs-Sigma specs

Context window
4K tokens
Max output
4K tokens
Mixture of experts
No
Parameters
8B and 3B (separate models)
Architecture
Decoder-only Transformer with a continuous-diffusion generation head

Training data

The card describes curated internal English text (webpages, articles, dialogue, code) of about 1.3 trillion tokens, split between autoregressive initialization and diffusion training. The data is not distributed.

Weight formats

  • safetensors

API endpoints

NameEndpointAuth
Hugging Face (8B base weights)https://huggingface.co/nvidia/Nemotron-Labs-Sigma-8BNone (open weights download)
Hugging Face (8B instruct weights)https://huggingface.co/nvidia/Nemotron-Labs-Sigma-8B-instructNone (open weights download)
Hugging Face (3B base weights)https://huggingface.co/nvidia/Nemotron-Labs-Sigma-3BNone (open weights download)
Hugging Face (3B instruct weights)https://huggingface.co/nvidia/Nemotron-Labs-Sigma-3B-instructNone (open weights download)

SDKs

  • Python

Regions

  • Global

Prompting patterns

Show 1 prompting pattern
  • Sample with DDIM at 32 steps per block

    sampler=ddim, steps=32, block_length=32, guidance_scale=3.0, score_temp=0.3

    The card recommends these settings on the log_snr grid, with stronger score temperature as the step count falls.

Known quirks

Show 3 known quirks
  • Base model is not a chat assistant

    The card says the 8B base is a pretrained starting point, best used with few-shot or benchmark-style prompts.

    Workaround: Use the instruct variant for chat and instruction following.

  • Quality drops below 16 steps

    The card says the pretrained models lose quality noticeably below roughly 16 denoising steps.

    Workaround: Switch to the distilled variants, which are trained for a fixed step count.

  • Loading runs vendor code

    Transformers loading needs trust_remote_code, which executes inference code stored in the repository.

    Workaround: Read the repository's code before enabling trust_remote_code on a shared machine.

Nemotron-Labs-Sigma facts

Deployment
Self-hosted
Trains on your data
No
Released
12 Nov 2026
Current version
v1.0
Family
Nemotron Labs
Variant
Base
Licence
NVIDIA Nemotron Open Model License · Open weights
Data retention
NVIDIA hosts no service for this model, so it receives and retains no prompts.

Data governance

Trains on inputs
No

Version history

  • v1.0 · 12 Nov 2026

    Hugging Face release date given on the model card; the paper was first submitted on 2 October 2026 and revised on 12 November 2026.

What changed

  • Release Nemotron-Labs-Sigma released on Hugging Face (v1.0) source

Questions about Nemotron-Labs-Sigma

How much does Nemotron-Labs-Sigma cost?

Nothing to download and no per-token price, because NVIDIA hosts no service for these models. Your own hardware is the only bill, and NVIDIA's speed test used one H100 in bfloat16.

Is Nemotron-Labs-Sigma open source, and can I use it commercially?

The weights are open under NVIDIA's own Nemotron Open Model License, which Hugging Face tags as 'other' rather than a standard open-source licence. The card says the model is ready for commercial or non-commercial use. NVIDIA does not distribute the training data or training code, so read the licence text before shipping a product.

Can Nemotron-Labs-Sigma run on a CPU?

No. The card says the diffusion sampler is numerically degraded on CPU and produces unusable output. It lists Linux and NVIDIA Ampere, Hopper, Lovelace or Blackwell GPUs as supported, with bfloat16 as the working precision.

Is Nemotron-Labs-Sigma better than Qwen3-8B?

In the paper's 8B table, Sigma reaches HumanEval 65.12 and GSM8K 88.57 in diffusion mode, against 64.63 and 86.73 for Qwen3-8B in autoregressive mode. Qwen3-8B wins MBPP, 69.40 against 65.74, while Sigma wins Minerva Math, 66.72 against 52.94. All figures are reported by NVIDIA's authors, and no independent board lists the model.

What sampler settings does Nemotron-Labs-Sigma need?

The card recommends DDIM on the log_snr grid with 32-token blocks: guidance 3 and score temperature 0.3 at 32 steps, 0.4 at 16 steps and 0.6 at 8 steps, then guidance 2 with 0.8 at 4 steps. DDPM degenerates at low steps, and below roughly 16 steps the distilled variants are the better choice.

About Nemotron-Labs-Sigma

Nemotron-Labs-Sigma is a family of continuous diffusion language models from NVIDIA, released on 12 November 2026 in 8-billion and 3-billion-parameter sizes. Instead of generating one token at a time, the model denoises Gaussian noise in a 16-dimensional token-embedding space and decodes text in parallel blocks, reusing finished tokens through a standard key-value cache. The accompanying paper, arXiv 2610.02665 (Large Language Continuous Diffusion Models), calls it the first family of large-scale continuous diffusion language models. NVIDIA's other open models on Hokai include the chat model Nemotron Labs Dory 1.0 4B and the autoregressive Nemotron 3 Nano 30B A3B, which work differently.

Read the full background

NVIDIA publishes 14 repositories on Hugging Face: base and instruct models at both sizes, distilled variants fixed at 2, 4, 8 or 16 denoising steps, and distill-spec variants built for self-speculative decoding. The base model is meant for research, fine-tuning and distillation, not as an assistant, and the instruct variants cover chat.

In the paper's tables, the DDPM sampler at 256 steps per block gives the 8B base model HumanEval 65.12 and GSM8K 88.57, averaged over 10 seeds. These are vendor-reported figures, since the authors are NVIDIA researchers, read from the paper on 17 November 2026. The same table gives the earlier Nemotron-Labs-Diffusion-8B 62.80 and 87.26 in diffusion mode, so the margin is small. Moving to the faster DDIM sampler at 32 steps per block changes the six-task average from 71.16 to 70.18.

Speed is the point of the design. NVIDIA's card measures roughly 120 to 140 tokens per second for 256 generated tokens on one H100 in bfloat16, using DDPM at 32 steps per 64-token block. The paper reports about 8x acceleration on math and 4x on coding from distillation, at 8 and 4 tokens per forward pass, with negligible accuracy loss.

The constraints are tight. The context window is 4,096 tokens, prompt and output together, and the card lists English as the only supported language. The model needs an NVIDIA GPU with CUDA and bfloat16 support on Linux, and it loads through Transformers with trust_remote_code. NVIDIA hosts no service for it, so it runs offline on your own hardware.

The card says the licence allows commercial or non-commercial use, and the weights sit in NVIDIA's Hugging Face organisation.

Production assistants, long documents and hosted APIs are the wrong fit, because the base model is not instruction-tuned, the window is 4,096 tokens and no endpoint exists. Teams that want diffusion generation through an API can look at Mercury 3, teams that want an open diffusion model with a much longer window can look at DiffusionGemma 26B-A4B, and an ordinary autoregressive 8B baseline is Ministral 3 8B.

NVIDIA says the models were trained on about 1.3 trillion tokens of internal English text: roughly 1 trillion for autoregressive initialization, then about 0.3 trillion of continuous-diffusion training. The privacy subcard says the datasets included personal data and were reviewed before release, and the training data and code are not distributed.

Sources and guides

More AI Models on Hokai

  • Hy4 flashTencent's Hy4 flash, released 16 November 2026, is a lower-cost multimodal model that reads text and images across a 1M-token window.
  • Qwen4 MaxAlibaba's Qwen4 Max, released 16 November 2026, is a sparse mixture-of-experts model with a 1M-token context and text, image and video input.
  • Gemini 4 HydrogenGoogle DeepMind's November 2026 Gemini 4 model for coding agents, with four thinking levels from low to xhigh and a free tier in AI Studio.
  • Qwen4-Max-PreviewAlibaba's November 2026 Qwen 4 flagship preview, open only to Token Plan subscribers, with three reasoning levels from low to xhigh.
  • Claude Fable 5Superseded by Fable 5.1 and Fable 5.2 but still sold on the Claude API: Anthropic's first generally available Mythos-class model, released June 2026, with a 1M-token context.
  • Claude Fable 5.1Superseded by Claude Fable 5.2 in November 2026 but still sold on the Claude API: Anthropic's second Mythos-class model, released Sept 1, 2026, for long-horizon coding and research.