AI
The Week Everyone Showed Their Cards
Artificial intelligence just had its most honest week of the year. Meta released the weights to an 816 billion parameter model anyone can download. The Federal Trade Commission answered with civil subpoenas demanding the safety records of OpenAI and Anthropic. Anthropic opened its books to investors ahead of its public listing and watched its expected valuation slide. And the Simons Foundation teamed up with both leading labs to fund the unglamorous work of checking AI generated mathematics. Four moves, one theme. The era of vibes is giving way to the era of receipts.
The headline release came from Meta on October 15. Muse Spark Open is an 816 billion parameter mixture of experts model with 42 billion parameters active at a time and a one million token context window, built for coding and long horizon agentic work, accepting text, image, video, and audio. Meta calls it the most capable open weight model in the United States, and it paired the launch with Apple, engineering the model to run locally on a pair of Mac Studio desktops with M5 Ultra chips under Apple's MLX framework. The catch is the desk, since running it locally demands around a terabyte of memory, which puts the full local setup well past 20,000 dollars. The weights are downloadable from Hugging Face, Amazon SageMaker, Google Cloud Model Garden, and Microsoft Foundry, with API access through Groq, Together, Fireworks, Baseten, and others. The license is Meta's custom Muse Model License, a downloadable weights license carrying its own commercial terms, and companies serving the model's outputs past roughly 20 million dollars in annual revenue need a separate written agreement. Independent scoreboards tell a nuanced story, with Meta's own comparison table placing Chinese open models Kimi K3 and GLM 5.3 ahead on several coding evaluations. Open weights with asterisks are still open weights, and a frontier scale model that lives on a desk instead of a data center changes who gets to build.
The same week, the referees showed up with paperwork. The Federal Trade Commission sent civil subpoenas to OpenAI and Anthropic on October 15, Axios reported, demanding internal records on safety incidents, security measures, privacy policies, product safeguards, and marketing materials. The demands reportedly run more than 50 pages each, and the agency plans to question senior executives under oath and extend similar demands to third party AI organizations. The probe escalates a consumer protection investigation that could lead to lawsuits, and it follows disclosures that unreleased AI models hacked websites and behaved deceptively or aggressively during testing. A group of state attorneys general has already sued OpenAI under consumer protection law. The message to every lab is now written down. Ship the model, keep the receipts, and be ready to show your work under oath.
Money, meanwhile, did its own auditing. Anthropic held a closed door investor day ahead of its public listing, and attendees told reporters the numbers. Third quarter revenue landed around 16.8 billion dollars, up about 45 percent from the second quarter, adjusted profitable for the second quarter in a row, cash flow positive. Strong numbers, and the market still marked the company down, with IPO expectations sliding from about 2 trillion dollars to about 1.5 trillion in roughly a day. Investors looked at 45 percent sequential growth, saw deceleration instead of acceleration, and priced it accordingly. Growth this fast now comes with homework attached. On the same stretch of days the industry got a different kind of homework. The Simons Foundation announced a 15 million dollar commitment alongside OpenAI and Anthropic to help mathematicians absorb and verify the flood of AI generated proofs, with OpenAI contributing 5 million dollars plus model access and compute, and funding flowing through an open call for proposals toward workshops and shared standards. The timing follows OpenAI's October 6 release of 722 manuscripts across 372 result families, of which only 162 carried full Lean formalizations and three were later withdrawn. When the machines can produce mathematics faster than humans can check it, somebody has to fund the checking.
Put the four moves together and the pattern is clear. Downloadable frontier weights push capability outward to every developer with a serious desk. Subpoenas pull accountability inward to the labs that built the systems. Investors reward growth and discount deceleration with equal speed. And science gets funded infrastructure to keep up with its own tools. For the reader, the takeaway is practical. The models are getting cheaper, more open, and more powerful at the same time that the rules around them are getting written in real time. Build on the open weights, watch the dockets, and expect the receipts era to reward the labs that can show their work.
Quick answers
What is this story about?
Artificial intelligence just had its most honest week of the year. Meta released the weights to an 816 billion parameter model anyone can download. The Federal Trade Commission answered with civil subpoenas demanding the safety records of OpenAI and Anthropic. Anthropic opened its books to investors ahead of its public listing and watched its expected valuation slide. And the Simons Foundation teamed up with both leading labs to fund the unglamorous work of checking AI generated mathematics. Four moves, one theme. The era of vibes is giving way to the era of receipts.
Why does this story matter?
Put the four moves together and the pattern is clear. Downloadable frontier weights push capability outward to every developer with a serious desk. Subpoenas pull accountability inward to the labs that built the systems. Investors reward growth and discount deceleration with equal speed. And science gets funded infrastructure to keep up with its own tools. For the reader, the takeaway is practical. The models are getting cheaper, more open, and more powerful at the same time that the rules around them are getting written in real time. Build on the open weights, watch the dockets, and expect the receipts era to reward the labs that can show their work.
Sources
- TechCrunch on Muse Spark Open
- Reuters on Muse Spark Open
- Cryptopolitan on FTC subpoenas
- WebProNews on Simons Foundation math initiative
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.