We took a 30B model and split it in... two to write tokens in parallel instead of one at a time. Introducing Nemotron-Labs-TwoTower: a diffusion language model from NVIDIA Research adapted from Nemotron-3-Nano-30B-A3B. Here’s how it works: one half holds the context, the other writes the tokens, with both reusing the pretrained model instead of training a new one from scratch. We found it kept 98.7% of the original model’s quality at 2.42× faster generation.show more

Added context to my tiny diffusion model to enable... sequential generation of longer outputs! Currently the context is a quarter of the sequence length (seq_len=256, context_len=64). I have a theory that the less semantic-value-per-token, the worse the “curse of parallel decoding” is. With parallel decoding, we independently predict multiple tokens in one step. With the sentence “My poker hand was a ___ ___”, two valid predictions are “two pair” and “straight flush”. Because each token prediction is independent though, we can end up with a nonsensical output like “two flush”. This seems to be exacerbated with low semantic-value-per-token, as now you need more tokens to express the same concept. Instead of needing to independently predict two tokens, we might need to predict 10 instead (which is of course much harder). The model currently has noticeably worse output compared to nanogpt (similar size) and I believe this is a main reason. I’ll try adding confidence-aware parallel decoding (from NVIDIA’s Fast-dLLM paper) and other tricks and see how much they improve generation quality.show more

4/ to achieve maximum memory efficiency, we quantize model... weights to ~4-bit, getting the language model under 20GB with room for the kv cache, perception encoder, and drafter alongside it. a dflash drafter proposes blocks of tokens the main model verifies in parallel, so it stays responsive.show more

Selected as a best paper finalist at #CVPR2026: PixelDiT... from NVIDIA Research In most image generation models, a pretrained autoencoder compresses the image before any diffusion happens, causing quality loss that accumulates across the entire pipeline. PixelDiT, or Pixel Diffusion Transformers, removes this step entirely. It's a single-stage model that learns the diffusion process directly in pixel space, end-to-end.show more

PAYING PER MODEL IS THE DUMBEST THING IN TECH... RIGHT NOW i was paying 3x what i needed to for AI inference the grid lets you buy a quality spec instead of a specific model.. it routes every request in real time to the cheapest option that qualifies swap one url and your code keeps working exactly the same openai-compatible, one line to switch, 200M free tokens to startshow more

Diffusion for virtual worlds: we've trained a new model... to create 3D objects from text. And it's 50x faster than any alternative. We made each of these cute characters in just one minute with Genmo's new text-to-3D model. Build your own world atshow more

the fact that i can take an image of... a room and turn it into a 3d model in one shot is actually insane this took like 30 seconds from image to 3d modelshow more

A peanut-sized Chinese model just dethroned Gemini at reading... documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.show more

A peanut-sized Chinese model just dethroned Gemini at reading... documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.show more

how small can you make a world model? we... built mira mini, the first reproduction of mira: •⁠ ⁠reproduced in a week from scratch •⁠ ⁠four players, one shared rocket league match, in the browser •⁠ ⁠distilled to 364m params, weights, code, full report: all open. play it here ↓show more

I MADE MY AI AGENT 10X FASTER WITHOUT CHANGING... THE MODEL not a smarter model, not a bigger context window, not another clever prompt the same kind of AI that designs vaccines for viruses we have not even met yet was spending two minutes opening the wrong files just to hand me a brief from three months ago the problem was never capability, it was the scaffolding that piled up around my agent by accident, folder by folder an agent does not think in your categories, it searches from scratch every single time, and your tidy human folders are a maze to it the fix was almost stupidly small, one index file at the root of each big folder and a few numbers in front of the folder names slowest task dropped from 2 minutes to 26 seconds, fastest ones hit 10, zero model changes capability is cheap when the scaffolding around it is broken the article breaks down the whole system in 15 minutes ↓show more

Nemotron 3 Ultra performed GPT 5.5 level 10× cheaper... We gave three same prompts to build HTML5 canvas with real physics. At first scene we have water in a spinning drum. Galton board - balls through pegs into bins. And a block collision setup with extreme mass differences. Outputs: Nemotron 3 Ultra: 11.3k tokens, $0.051 GPT 5.5: 11.0k tokens, $0.57 Nemotron stays right on GPT 5.5's heels, but at 10× cheaper. The gap in quality is far smaller than the gap in price.show more

NVIDIA just made AI detect objects 10x faster by... deleting one step. It's called LocateAnything, and it removes the biggest bottleneck no one else was fixing in vision-language models. Normally a model builds each bounding box one coordinate token at a time. 100 objects means thousands of tokens before an answer. NVIDIA scrapped that: their Parallel Box Decoding predicts the whole box in a single forward pass, as one atomic unit. → 12.7 boxes/sec on one H100 → 10x faster than Qwen3-VL → +3.8% F1 on LVIS, accuracy up, not down → 3B params, runs on one consumer GPU Treating the box as one unit keeps its coordinates tied together, which is why accuracy climbed instead of falling. One model handles detection, GUI grounding, OCR, and document understanding, ready for computer-use agents, robotics, and document pipelines. 100% open source, weights, code, demo, and paper all live.show more

i gave Ox Alpha the main menu from ECHO... and asked it to recreate it only ONE (1) prompt was enough to build this, using the reference this is one of the most beautiful output i've ever had from any model the fact that it's from a stealth model is mindblowingshow more

🚀 Self-speculation brings 6.75x real speedup for LLM generation... with SGLang inference! Same model drafts future tokens in Diffusion mode → then verifies them in AR (causal) mode. One model and one KV cache. Just different attention masks. Thanks to perfect alignment, we get 2× longer acceptance lengths than MTP techniques (Eagle-3, MTP, dFlash). We run 2 forward passes… but the 2× higher acceptance means we break even - and with zero overhead from extra drafter, KV cache, or LM head that comes with MTP - those are not free. Last week we released Nemotron-Labs-Diffusion + Tri-mode LLMs! We did continued pre-training on Ministral-3 models by switching attention patterns (block causal bidirectional). Result: one model that runs AR mode, Diffusion mode, and Self-Speculation. Diffusion mode already shows high benchmark accuracy - excited to see what happens when someone beats left-to-right acceptance! 🔥 Github: Paper: SGLang inference: Try the models on HF:show more

You can now vibe code a language model. From... a single prompt, GPT‑5.6 built the entire training pipeline and trained a model from scratch on my iMessage history. Locally on my Mac. It now generates replies in my writing style.show more

One of the top pieces of feedback we get... about OpenAI Codex: “make it faster”. We addressed it in a big way with our ultra fast Codex-Spark model (research preview for Pro). Available in the latest Codex app, CLI, and IDE extension. Here’s Spark vs 5.3-Codex-Low side-by-side (Credit Romain Huet for the initial idea).show more

Soon we will launch our new model at MagicPath.... It will be the best model in the world for interface design. Design is, and always will be, the moat. Here are some one-shot results from an early checkpoint.show more

1/ Happy to share UniDisc - Unified Multimodal Discrete... Diffusion – We train a 1.5 billion parameter transformer model from scratch on 250 million image/caption pairs using a **discrete diffusion objective**. Our model has all the benefits of diffusion models but now in multimodal space! - flexible compute-quality tradeoff, zero-shot inpainting and editing, better control via classifier-free guidance and lower latency! We open source everything - our code, weights and the training dataset.show more

Seedance 2.0 has a fix for character drift almost... nobody uses. Every new text description is a fresh interpretation of appearance. That's where the drift comes from: a slightly different face, a slightly different outfit, proportions that shift from scene to scene. The @ Image tag solves this directly — attach a photo once, and it becomes the fixed source of truth for face, outfit, and proportions across the entire multi-shot sequence. The @ Video tag works the same way for motion and camera — tag a reference instead of describing it, and the model copies that exact style instead of an approximation. You're not making the model remember. You're just stopping yourself from reminding it differently every single time.show more