WORLDTECH NEWS Global technology intelligence.Contact
AI SINGLE SOURCE

NVIDIA Releases Alpamayo 2 Edge: A 2B Reasoning VLA for DRIVE Thor

Close-up of two NVIDIA RTX 2080 graphics cards with dual fans, high-performance hardware.
Illustrative photo.Photo by Nana Dua on PexelsNVIDIA logo shown for identification only; no affiliation with or endorsement of WORLDTECH is implied.

What happened

NVIDIA released Alpamayo 2 Edge, a 2B-class reasoning vision-language-action model that plans driving trajectories and explains them on DRIVE AGX Thor, at 160 to 190 ms pixel-to-trajectory latency (the delay between asking for something and getting it). NVIDIA has released Alpamayo 2 Edge , a compact reasoning vision-language-action (VLA) model for autonomous driving. NVIDIA is a semiconductor company based in Santa Clara, and its products and services include graphics processing units.

Context length: not disclosed (measured prompt: 3,236 tokens). Performance: Strong closed-loop driver for its size; mid-pack on reasoning QA.

Alpamayo 2 Edge is a 2B-parameter model that plans trajectories and explains them inside the vehicle. It targets NVIDIA DRIVE AGX Thor , not a datacenter GPU (a chip built for many calculations at once, used for graphics and AI).

The goal: reasoning-based planning within onboard compute and latency budgets. Card breakdown: 2.45B backbone + 0.2B action expert (HF metadata says 3B).

Runs on: NVIDIA DRIVE AGX Thor with the FP8 + TensorRT Edge-LLM (the kind of AI system trained on text to produce text) recipe. BF16 PyTorch needs 16 GB+ GPU memory, validated on 1x H100 80GB.

Worst: nuReasoning numeric score of 48.67; LingoQA 75.2 trails Alpamayo 2 Super (79.2). Worst: only H100 validated, and real-time Thor speed needs the FP8 recipe.

Key facts

  • NVIDIA — released: Alpamayo 2 Edge, a 2B-class reasoning vision-language-action model that plans driving trajectories and explains them on DRIVE AGX Thor, at 160 to 190 ms pixel-to-trajectory latency
  • NVIDIA has — released: Alpamayo 2 Edge , a compact reasoning vision-language-action (VLA) model for autonomous driving

Sources & evidence