Nvidia launches Alpamayo 2 Super for commercial robotaxi use
Nvidia has made its Alpamayo 2 Super model publicly available for commercial robotaxi and autonomous‑vehicle development. The model is a 34‑billion‑parameter vision‑language‑action system built on the Cosmos 3 Super Reasoner and post‑trained with reinforcement learning. It is released under the Linux Foundation’s OpenMDW‑1.1 license, allowing fine‑tuning, derivative models and commercial redistribution, and the weights are hosted on Hugging Face.
Alpamayo 2 Super processes video from up to seven cameras to provide 360‑degree surround perception. For each driving scenario it outputs a planned trajectory, a chain‑of‑causation trace that explains the decision, meta‑actions such as yielding or lane‑changing, visual question‑answering linked to image regions, and automatically generated training labels. Nvidia reports that the model tops the LingoQA benchmark, beating rivals such as GPT‑4o, Gemini 2.5 Pro and Qwen2.5‑VL by 15‑23 points.
The release is positioned as a step toward more interpretable autonomy, helping developers address rare edge cases and easing regulatory certification. Nvidia senior director Marco Pavone and CEO Jensen Huang highlighted the model’s potential to accelerate robotaxi deployment while Nvidia continues to sell its DRIVE hardware for the distilled, in‑vehicle versions of the model.
Entities: Alpamayo 2 Super · Cosmos 3 Super Reasoner · Hugging Face · Jensen Huang · Marco Pavone · Nvidia Corp · Nvidia Corp.
Claims
What the coverage asserts, and how well corroborated each claim is across sources.
- [● 3 SOURCES] The model provides 360‑degree surround perception using up to seven vehicle‑mounted cameras. (NVIDIA)
- [● 2 SOURCES] The model is released under the Linux Foundation’s OpenMDW 1.1 license on Hugging Face, allowing commercial use, fine‑tuning and redistribution. (NVIDIA)
- [○ 1 SOURCE] The model has been downloaded over 500,000 times on Hugging Face and is used by major OEMs. (NVIDIA)
- [● 3 SOURCES] It can generate planned vehicle trajectories and chain‑of‑causation traces explaining its driving decisions. (NVIDIA)
- [● 3 SOURCES] NVIDIA released Alpamayo 2 Super, a 34‑billion‑parameter vision‑language‑action model for autonomous vehicles. (NVIDIA)
- [○ 1 SOURCE] The open‑source release can reduce manual annotation time for rare driving scenarios from months to days. (NVIDIA)
- [○ 1 SOURCE] In NVIDIA’s internal LingoQA benchmark the model achieved a score of 79.2, ranking first among roughly 40 models and outperforming Qwen2.5 VL 72B, Gemini 2.5 Pro and GPT‑4o. (NVIDIA)
- [○ 1 SOURCE] Training used approximately 115,000 hours of multi‑camera driving video and about 3.7 million chain‑of‑causation reasoning traces. (NVIDIA)