Wispr Flow - Don’t type, just speak
Mini toolsNVIDIA Alpamayo 2 Super
By NVIDIA
Alpamayo 2 Super extends the Alpamayo family from 10B to 32B parameters, improving reasoning, 3D spatial understanding, and trajectory prediction in long-tail driving scenarios. Built on NVIDIA Cosmos world foundation models, it adds full-surround 360-degree perception across front, side, and rear cameras. Key capabilities include Meta-Action outputs (yield, lane change, stop), reasoning auto-labeling with 2D grounding that compresses annotation cycles from months to days, and improved chain-of-causation (CoC) traces. Designed as a teacher model, it can be distilled into smaller models deployable on NVIDIA DRIVE AGX Thor in-vehicle compute. Supports multitask capabilities including scene understanding, model critiquing, and knowledge distillation. Inference code is available on GitHub and model weights on HuggingFace. Developers can use post-training scripts to adapt the model to custom datasets, scenarios, and driving policies for scalable L4 AV development.
New
Multimodal
Gen 3
Overview
A 32-billion-parameter open reasoning-based vision language action (VLA) model for level 4 autonomous vehicle development. Processes 360-degree multi-camera input to enable perception, reasoning, and trajectory planning across the full driving stack. Outputs chain-of-causation traces, Meta-Actions, and trajectory predictions.
About NVIDIA
Industry:
Computer Hardware Manufacturing
Company Size:
42000
Location:
Santa Clara, California, US
Website:
nvidia.com
Last updated: June 23, 2026
