Today we're shipping Nemotron 3 Ultra. A 550B MoE frontier-intelligence open model built for long-running agents. It delivers 5x faster inference and lowers the cost of complex agentic tasks by up to 30% versus other open frontier models.

People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way. We share our approach, early results, and a quick look at our model in action. thinkingmachines.ai/blog/int…

New preprint from @lightningrodai! We trained AI to predict clinical events — ICU transfers, new diagnoses, complications, procedures, ventilation, mortality — directly from raw clinical notes. No labeled data required – Foresight Learning infers outcomes from what happens later in patient records. Using Tinker from @thinkymachines , we trained a lightweight adapter on GPT-OSS-120B, resulting in a specialized predictor that runs on a single GPU. Results: 🎯 ~70% lower calibration error 📈 Brier skill score: ~0% → 27% 🧠 84% win-rate vs the base model in blind reasoning review 🥇 Slightly better Brier than GPT-5, despite being a fraction of the size Hospitals and specialty clinics often treat unique patient populations that out-of-the-box models don't have training data for. This makes it possible to build frontier-quality predictors for highly specific patient groups, with nothing but raw clinical records. Congrats to the team — @indiequant @KSkotheim64001 🙌 Full paper 👇 arxiv.org/abs/2605.12817

How does Exa compare to Google for training LLMs to search? In this blog post, we find that LLMs using Exa during reinforcement learning reach higher performance with 70% less training compute. exa.ai/blog/rl-search-outcom…

the @tinkerapi tutorials were really well put together thank you @thinkymachines folks this was really helpful for the project I'm working on

In Agent RL, models suffer from Template Collapse. They generate vast, diverse outputs (High Entropy) that lose all meaningful connection to the input prompt (Low Mutual Information). In other words, agent learn different ways to say nothing. 🚀 Introducing RAGEN-v2 -- Here's how we define and fix such silent failure modes in Agent RL. 🧵

FrontierSWE was built with collaborators from industry and academia to ensure that tasks are diverse and reflect real work engineers and researchers encounter. We specifically thank our partners @Modular, @PrimeIntellect and @thoughtfullab for their contributions

Another task tests AI research capabilities: using @tinkerapi from @thinkymachines, agents are asked to post-train an agent to play logic games, which involves writing an entire training pipeline and running experiments with different recipes to finally submit the best model

Introducing FrontierSWE, an ultra-long horizon coding benchmark. We test agents on some of the hardest technical tasks like optimizing a video rendering library or training a model to predict the quantum properties of molecules. Despite having 20 hours, they rarely succeed

I pointed Claude Code at a research task (build a golf forecasting system) and let it run for 49 hours on Tinker. No human in the loop. It ran 108 experiments. Here's the full trajectory, including the ones that made things worse.

We’ve redesigned our docs with easy access to SDK reference, tutorials, support, and our newly updated cookbook---v0.3.0! Whether you’re writing your first training loop in Tinker or debugging async RL, we want to make it easier to find what you need.

First, to get you started, we've created 23 tutorials to walk you from the API basics to advanced training techniques and deploying models into production. tinker-docs.thinkingmachines…

I know it's self serving to say, but man I would've killed for a resource like Tinker and the tutorials, the cookbook, etc back when I was in undergrad. Following @karpathy blogs and training RNNs on a crappy Acer *was* fun, but doing bigger things with less setup is such a boon