Today, we’re making Praxis-1 available to everyone. Our state-of-the-art world action model turns video pretraining into control for real robots, and it ships with open weights under Apache 2.0, ready to run and fine-tune on your own hardware.
Software engineering is evolving in the age of AI. One way to conceptualize it:
We shift from building code, to building the machines that build the code. And in doing so, we won't evaluate the code, we will evaluate the machines.
Human attention transitions from PR-level change review to continuous system optimization.
If you’re comparing models by their price sheet, you’re missing half the math.
Jacob Loewenstein @spatialjlo from NVIDIA explains why token efficiency belongs at the top of your evaluation criteria, whether you’re adopting a model or training your own.
One of the standout moments from Open by Design NYC, where Jacob, Zach Wentz @zkwentz from Reflection, and Hassan El Mghari @nutlope from Together AI chatted with builders digging into open models, post-training, and inference.
Tired of waiting for your diffusion models?
🚢 We just shipped a major FastGen update:
- new models: H3, LTX-2.5, LingBot-World, Cosmos 3, FLUX.2
- new methods: PDD, AYF, AnyFlow, Causal Forcing
- new features: multimodal distillation, image editing, camera conditioning, KV cache management, ComfyUI
🔗 Try it out here: github.com/NVlabs/fastgen
(1/n) We’re sharing Nemotron-Labs-Diffusion-Uni: a unified diffusion language model for multimodal understanding and generation. It achieves SoTA understanding and generation performance by learning from autoregressive VLMs and continuous flow models via On-Policy Distillation
We're looking to hire DevRels for Dynamo!
Awesome parts about the job:
- Getting to be a community face for one of the most important domains in the world (Inference)
- High visibility
- Extremely motivated and fast-moving team
- Working in open source
- Lots of compute
/more
Today we’re releasing two decision models, elephant-1 (9B) and mammoth-1 (27B), trained by the @baselabs team, available for free for one week on our Model APIs.
mammoth-1 is #1 on the Jev Decision Index and elephant-1 is tied for #1 on the Artificial Analysis Decision Index.
It’s a new model, but more importantly, this is an open-source model you can access, deploy, and post-train on Baseten (or wherever you do your best work).
The free week ends 10/27/26 at 11:59 PM PT. Try it and let us know what you think.
Try mammoth-1 here: baseten.co/library/mammoth-1…
Try elephant-1 here:
baseten.co/library/elephant-…
Over the past two years of building @_inception_ai, one question has come up for me more than any other across countless conversations: Does the diffusion advantage scale?
Today, we're answering that question with Mercury 3, our most capable diffusion LLM yet. Mercury 3 surpasses GPT-6-Luna and Claude 5.5 Haiku while delivering blazing-fast inference at 1000+ tokens/sec.
More importantly, the Mercury factory is up and running: Compared to Mercury 1, Mercury 3 is 40x larger and 3x more intelligent, all while maintaining the same industry-leading speed.
The diffusion advantage doesn't just scale. It compounds.
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
2/4 Mercury 3 demonstrates that text diffusion scales. At over 40x the size of Mercury 1, it’s vastly more intelligent while remaining just as fast. This is just the beginning: diffusion opens a new path to scaling AI, where intelligence can grow without a proportional increase in latency.
dLLMs remain one of the coolest and most slept on architecturally developers. Big leap fwd in intelligence and you can't get this kind of speed anywhere else w/o hardware acceleration!!!
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
Beam weights are out today (Apache 2.0 license).
We integrated Beam across 40+ different partners.
And are also open sourcing Mirror CLI - the harness Beam was trained against.
Happy building.
Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners.
We're releasing:
- Weights, tokenizer, and reference implementation
- BF16, FP8 and NVFP4 weights and a DFlash2 drafter
- The open source harness we built and trained Beam on: Mirror CLI
Weights and model card on Hugging Face: huggingface.co/reflection
Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners.
We're releasing:
- Weights, tokenizer, and reference implementation
- BF16, FP8 and NVFP4 weights and a DFlash2 drafter
- The open source harness we built and trained Beam on: Mirror CLI
Weights and model card on Hugging Face: huggingface.co/reflection
Incredibly proud of @reflection_ai for immediately jumping into the #2 spot for US open models with their first ever model. Beam is SO fast and efficient for its size. Check it out
Reflection’s first model, Beam, scores 30 on the Artificial Analysis Intelligence Index, placing it #2 among US open weights models
Beam is an open weights model with 501B total and 23B active parameters, released on @huggingface
Congratulations to @reflection_ai on the release!
Key takeaways:
➤ Second among US open weights models: Beam scores 30, behind only Meta's Muse Spark Open (39) and ahead of yesterday’s NVIDIA Nemotron 3.5 Super VL (26) release. Outside China, only Muse Spark Open and the UAE's K2 Horizon (31) score higher
➤ Token efficient for its intelligence: Beam uses ~26k output tokens per Intelligence Index task compared to K2 Horizon, one point ahead, which uses twice as many (~52k). Proprietary models still set the frontier, with GPT-6 Astra (max) scoring 53 on the same budget (~27k)
➤ Strong on agentic workflows, weaker in the terminal: Beam scores 48.5% on AutomationBench-AA, 11 percentage points ahead of K2 Horizon (37.2%) and level with Qwen3.8 27B (48.2%). Terminal-Bench 4.0 is a weak point, where Beam scores 3.0%, behind Qwen3.8 27B (5.6%) and Muse Spark Open (8.1%)
➤ Knows more, guesses more: On AA-Omniscience it answers 33% of questions correctly, well ahead of K2 Horizon (18%) and Qwen3.8 27B (16%). But its hallucination rate is 47%, against 26% for K2 Horizon and 30% for Qwen3.8 27B, giving Beam a net index score of 1.5, ahead of both peers (−3.0 and −10.0)
Other model details
Context window: 1M tokens
Modalities: Text
Parameters: 501B / 23B
Nvidia model strategy:
1. Build great base models
2. OpenSource model, weights, training data & pipelines
3. Engage an ecosystem of partners for post training & domain or task specific fine tuning
4. Make available everywhere at great economics
5. Solve real customer problems 🎉
Meet NVIDIA Nemotron 3.5 Super VL
Our new open model combines reasoning and tool use with text, image, document, and video understanding for long-running AI agents.
120B total parameters with 12B active 🧵