Plus: Apple widens its OpenAI trade secrets fight to 11 more ex-employees, and NVIDIA opens its top robotaxi model for commercial use.
MODELS
🧮 Astra cracks ten long-stuck math proofs while OpenAI stays quiet on the launch
Image source: gizmodo.com
OpenAI used a blog post about math proofs to introduce Astra, described as "our next major model." An internal version produced ten proofs covering problems stuck for at least a decade, at a reported cost of roughly $2,000 at Sol API rates. Whether Astra ships as GPT-5.7, GPT-6, or something else remains undecided.
The details:
OpenAI announced Astra, described as "our next major model," in the third paragraph of a blog post titled "Ten advances in mathematics and theoretical computer science."
An internal version of Astra produced ten advances in mathematics and theoretical computer science, with OpenAI saying the problems had seen no progress for at least a decade.
Astra formalized every argument as a Lean certificate so the proofs could be checked by the mathematical verification system, while human researchers used the same model to prepare the manuscripts.
The Information reported that Astra is built for "long-running" workloads and that Sam Altman demoed the model to federal officials in Washington, D.C. over the past week.
Why it matters: OpenAI just turned math proofs into a product demo, and the $2,000 price tag is the real headline — research-grade output at API rates changes who can afford serious AI work. The Washington demos over the past week suggest the audience that matters most is federal officials, not developers. Watch the naming decision: whether Astra ships as GPT-5.7 or GPT-6 signals how big OpenAI thinks this jump is.
⚖️ Apple widens its OpenAI trade secrets fight to 11 more ex-employees
Apple says its trade secrets investigation now reaches beyond Chang Liu and Tang Tan, with 11 other former Apple employees possibly witnesses or involved. OpenAI is pushing back hard, publishing a blog post titled 'Apple is getting this wrong' and calling the lawsuit careless, aggressive, and oddly personal.
The details:
Apple filed suit against OpenAI on July 10 in the US District Court for the Northern District of California, alleging misappropriation of trade secrets under the Defend Trade Secrets Act.
Apple seeks a preliminary injunction to stop Liu, Tan, and OpenAI from accessing, using, or disclosing alleged confidential information and to halt an AI device based on Apple's technology.
Apple's new filing requests expedited discovery and says its investigation revealed 11 other former Apple employees may have been witnesses or involved beyond Liu and Tang Tan.
OpenAI published a blog post titled 'Apple is getting this wrong,' sharing iMessage and email exchanges, and says Apple's lawyers emailed the wrong person after confusing two similar Asian surnames.
Why it matters: Apple is signaling this is bigger than two defectors: more than 400 former Apple employees now work at OpenAI, and Apple says it warned OpenAI back in February. OpenAI says it neither has nor wants Apple's trade secrets. Will a court grant expedited discovery? If so, hiring pipelines from Cupertino to OpenAI face real scrutiny, and every big-tech AI hardware team just got a legal precedent to watch.
🏛️ NVIDIA helps NSF take its Florida AI playbook national
NVIDIA is joining the U.S. National Science Foundation's State and Regional Artificial Intelligence Infrastructure Hubs program, launching today. The effort expands access to computing, data, software and expertise for AI research and education at state and multistate university groups, modeled partly on NVIDIA's 2020 partnership with cofounder Chris Malachowsky and the University of Florida.
The details:
NVIDIA is participating in the NSF State and Regional Artificial Intelligence Infrastructure Hubs program, launching today to expand access to computing, data, software and expertise for AI-enabled research and education.
The program will support state and multistate groups of colleges and universities, modeled partly on the 2020 public-private partnership among NVIDIA, cofounder Chris Malachowsky and the University of Florida.
Since launching its university-wide initiative in 2020, UF now counts more than 300 AI-focused faculty, with AI education and research across all 16 colleges and more than $511 million in research awards since 2017.
NVIDIA describes itself as a leading contributor to the NSF-led NAIRR pilot program, the National Artificial Intelligence Research Resource effort on which the announcement is built.
Why it matters: Washington gets a national AI research network without building it alone, and NVIDIA gets its computing and expertise embedded in public science infrastructure. The UF template shows the stakes: one partnership scaled to 300 AI faculty and $511 million in awards. Watch which states land hubs first — early winners will shape where AI talent and grant money concentrate for years.
🚕 NVIDIA opens its top robotaxi reasoning model for commercial use
Image source: blogs.nvidia.com
NVIDIA Alpamayo 2 Super is now available for commercial use on Hugging Face under the OpenMDW-1.1 license, which covers fine-tuning, derivative models and commercial redistribution. The model ranks first on the LingoQA autonomous driving reasoning benchmark among nearly 40 models evaluated, as the Alpamayo family surpasses 500,000 downloads on the platform.
The details:
NVIDIA Alpamayo 2 Super is now available for commercial use on Hugging Face under the OpenMDW-1.1 license, which covers fine-tuning, derivative models and commercial redistribution.
Alpamayo 2 Super ranks first on the LingoQA autonomous driving reasoning benchmark among nearly 40 models evaluated, producing five tightly coupled outputs per driving situation.
In NVIDIA testing using the Lingo-Judge metric, Alpamayo 2 Super outperformed Qwen2.5-VL 72B by 17.0 points, Gemini 2.5 Pro by 15.1 points and GPT-4o by 23.2 points.
The OpenMDW-1.1 license covers fine-tuning, derivative models and commercial redistribution, and is being applied across the entire Alpamayo model family.
Why it matters: NVIDIA is handing every robotaxi and AV startup a top-ranked reasoning model it can fine-tune and commercially redistribute, not just demo — cutting the build cost for teams without frontier-model budgets. The Lingo-Judge gaps, 23.2 points over GPT-4o and 15.1 over Gemini 2.5 Pro, give adopters a measurable edge. Watch whether OpenMDW-1.1, now applied across the entire Alpamayo family, becomes the default license for open driving models.
ChatGPT The AI assistant even Congress is expensing — if you haven't tried it for drafting, summarizing, and brainstorming yet, you're behind Capitol Hill.
Researched and written by AI Newsroom, by Classwise.io
