Fetching from the wire…
Public story · 2026-08-05 · high
The license covers Nvidia's entire Alpamayo family, not just this release, and opens the door to commercial fine-tunes.
Why now: Nvidia's blog post announcing the relicensing and Alpamayo 2 Super's release is dated August 4.
Nvidia relicensed the Alpamayo family of autonomous-driving reasoning models under the Linux Foundation's OpenMDW-1.1 on August 4. That's every version, not just the new one. The terms cover Alpamayo 2 Super, the roughly 30-billion-parameter flagship released the same day, and every earlier version in the family. Robotaxi teams can now fine-tune, modify, and ship commercial products on any of them under the new terms.
Alpamayo 2 Super is about three times the scale of the prior Alpamayo 1.5. It ranks first on LingoQA, a driving-reasoning benchmark, among roughly 40 models. On Lingo-Judge scoring it beats Qwen2.5-VL 72B by 17.0 points and Gemini 2.5 Pro by 15.1 points, per Nvidia's announcement.
A 30-billion-parameter model beating a 72-billion-parameter one on a driving benchmark is solid, but labs claim wins like that most months. A hardware company giving away commercial rights to its whole reasoning-model family is rarer. It's a bet that owning the layer robotaxi teams build on matters more than what they'd pay to license the model itself.
The relicensing landed alongside Alpamayo 2 Super's release on August 4, both folded into one announcement. Watch for the first commercial product built on a fine-tuned version of Alpamayo from outside Nvidia. That's the real test of whether the license mattered more than the benchmark.
Each link below shares sources, entities, or timing with this story.
Willison's August 2 roundup lays out "Open Weights and American AI Leadership" (July 24, Microsoft-shepherded, now 235 signatory companies including NVIDIA, Amazon, Y Combinator, the Linux Foundation, and OpenAI after initially abstaining); Anthropic's separate July 27 rebutta...
RFC published August 4 by an Open Secure AI Alliance working group spanning 120+ organizations including NVIDIA, Cisco, CrowdStrike, Hugging Face and Red Hat. SAFE proposes confidentially collecting agentic AI incidents, notifying impacted parties, identifying recurring contro...
The changelog offers Z.ai's open-weights coding model with a 1M-token context free via Blackbox AI on AI Gateway, default for new eve agents, switchable for existing ones with eve set --model zai/glm-5.2. Excludes Fast mode and the glm-5.2-fast variant. Separately, Gemini 3.7...
Stratechery's August 14 roundup lays out the position that requiring providers to mark all AI outputs as AI-generated is philosophically wrong, drawing a parallel to the history of idea propagation across centuries (Stratechery). Anthropic responded to the rule by watermarking...
Google shipped it August 13: FrontierCode 1.1 at 43.6% (from 34.4%), DeepSWE v1.1 at 65.3% (from 49.0%), WebDev Arena Elo 1588 (from 1538), AutomationBench enterprise workflow completion at 30.4% (from 17.0%). Introductory pricing is $0.75/M input and $3.75/M output through De...
Three moves, two days, no coordination between them. August 10–11: GitHub shipped Ollama as a BYOK provider inside Copilot for JetBrains (GitHub Changelog). Unsloth released Unsloth Desktop with a command literally named unsloth start claude, which points Claude Code and Codex...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.