New Models
2026-08-10
Nvidia just gave away the model that drives the car — 34 billion parameters, and you may sell what you build with it
Alpamayo 2 Super is now available for commercial use on Hugging Face. It is a roughly 34-billion-parameter vision-language-action model for autonomous driving: it takes camera input, reasons about the scene in text, and outputs a driving trajectory, plus the reasoning trace behind it and automatic labels for training data. It was built on Nvidia's Cosmos 3 Super Reasoner and then post-trained with reinforcement learning. The licence is the part that matters. Alpamayo 2 Super ships under OpenMDW-1.1, the Linux Foundation's permissive licence for open model distributions, which covers fine-tuning, derivative models and commercial redistribution — a company can distil it into something small enough to run in a car and sell that, without asking Nvidia. On LingoQA, a reasoning benchmark for driving, Nvidia's own testing places it first among roughly forty models evaluated, ahead of Qwen2.5-VL 72B by 17.0 points, Gemini 2.5 Pro by 15.1 and GPT-4o by 23.2. The Alpamayo family has passed 500,000 downloads on Hugging Face. The benchmark numbers come from the vendor and have not been independently reproduced.
Why it mattersSelf-driving has spent a decade as the most closed corner of AI: every serious stack was proprietary, every dataset was a moat, and a new entrant needed years and a fleet before it had anything to test. Nvidia has now put a competent driving brain on a public download page under a licence that lets you sell what you make from it. That is not charity. Nvidia sells the chips these models run on, and the fastest way to grow the market for driving computers is to remove the reason a startup, a bus manufacturer or a national robotaxi programme would give up before starting. Watch what the openness actually buys, though: reasoning traces mean the model writes down why it chose a manoeuvre, which is exactly the artefact a regulator or a crash investigator needs and has never been able to get out of a black-box driving stack. If open reasoning models become the norm here, accident inquiries in this field change character — from arguing about what the car saw to reading what the car said.
✓ Verified · 4 sources
▶ Related video: Introducing NVIDIA Alpamayo 2 Super
Read in the app — free, in 9 languages
Related stories
Nevada just cleared up to 8,000 driverless taxis for Las Vegas — Tesla's ceiling went from 10 to 5,000 in five weeks
2026-08-21DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best
2026-08-21The doctors and lawyers who grade AI answers are now a half-billion-dollar business — and this one grew fivefold in eight months
2026-08-21Show the robot once — three to twelve seconds — and it gets the job right 59 times out of 100 with no training at all
2026-08-21The robot dog China sells the world was measured, to the millimetre, from one the US Army paid to invent
2026-08-20