OpenAI published 722 mathematical manuscripts produced by an unreleased internal frontier model, organized into 372 result families and released after consultation with an independent advisory group at the Institute for Advanced Study. The company framed it as evidence that AI systems are beginning to make novel research contributions, with Sam Altman declaring a "new era of discovery." Reaction split between celebration and scrutiny of how much of the work is genuinely verified — François Chollet countered that math and code are exactly where RLVR pushes farthest, while other domains stay bottlenecked by human data.
Mistral launched a public preview of Mistral Large 4 — nicknamed "le Chonk" — a 1-trillion-parameter natively multimodal MoE model with 49 billion active parameters that the company calls the most powerful open-weight system outside China, with weights promised by the end of October. It debuted at #45 on the Code Arena WebDev leaderboard, a +304-point jump over Mistral Large 3, and Artificial Analysis scores it on par with DeepSeek V4.1 Flash and GPT-6 Luna.
On infrastructure and policy: filings show Broadcom will lend Anthropic up to $42 billion in convertible notes to fund a $125 billion TPU compute lease while simultaneously serving as its chip supplier, and Anthropic separately committed $100 million to its Claude Frontier Academy talent program. New York City councilmembers grilled lab executives over oversight gaps, and Elon Musk announced Grok Bot will route tasks to the best available back-end model — including Anthropic's Claude Opus 5.5, MidJourney, and Suno.
OpenAI published 722 manuscripts from an unreleased internal frontier model, covering 372 result families that group related papers. The release was made with advice from an independent mathematics advisory group at the Institute for Advanced Study and follows an earlier ten-proofs release. The model was handed roughly 4,000 problems, averaging about three hours of ChatGPT Pro-level compute per result.
Mistral launched a public preview of Mistral Large 4 ("le Chonk"), a 1T-parameter natively multimodal mixture-of-experts model with 49 billion active parameters, trained from scratch in its own European data centers. The company positions it as the strongest open-weight model outside China, with weights scheduled for release by the end of October 2026. Preview API access is live through Mistral Studio.
Anthropic's IPO filing confirms Broadcom simultaneously supplies its AI processors, leases equipment, lends up to $42 billion in convertible notes, and guarantees a $60 billion credit facility backing a five-year, $125.2 billion TPU lease starting in 2027. Bank of America, Citi, and Morgan Stanley have begun syndicating the debt, and BofA analysts estimate Broadcom's financing vehicle could support roughly $370 billion of senior debt at 20-gigawatt scale by mid-2029.
Anthropic is committing $100 million to Claude Frontier Academy, a program to train 10,000 "Frontier Deployed Engineers" by the end of 2027, starting with enterprise cohorts from its Claude Partner Network. The centerpiece is a 12-week residency modeled on medical training, where engineers lead a real Claude deployment at their own company. Anthropic frames talent, not technology, as the most pressing bottleneck in enterprise AI adoption.
Executives from Anthropic, OpenAI, Google, and Meta testified before the New York City Council as local regulators weigh oversight of AI deployment. Councilmembers singled out the federal government's light-touch approach after President Trump signed a voluntary accord with tech leaders, with one saying "the idea that artificial intelligence is going to self-regulate defies all reason." The hearing follows a summer of containment breaches at OpenAI that sharpened public scrutiny.
Selected posts from the edition, with links to the originals.
"We're releasing a broad range of new mathematical results produced by an internal frontier model. We've been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and [input on] how we release these results."
View post on X ↗"Important note regarding Grok @Bot: Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is most likely to give you the best outcome."
View post on X ↗"After a brief but institution-eroding slop science era, it increasingly looks like we are going to have two revolutions from AI: 1) Everything ever published will be re-read and re-judged in ways that human scientists never anticipated 2) Novel discoveries will start to come fast"
View post on X ↗"What if the jagged frontier is mainly math + code (which you can push arbitrarily far with RLVR), and everything else starts to plateau because it is still bottlenecked by human generated data? Model performance in non-verifiable areas has kept improving steadily, albeit much more slowly."
View post on X ↗"mathematicians: frontier labs need to publish real proofs so we can check them, not blog posts. OpenAI: fine. here's 722 papers"
View post on X ↗"Mistral Large 4 by @MistralAI just landed in the Code Arena: WebDev at #45. With 1534 pts, this is a +304 pt improvement from Mistral Large 3 at #130! This release also marks a 271-point improvement over its next-best-performing variant, Mistral Medium 3.5, at #122."
View post on X ↗"I hooked up our team claw to X to trigger work faster. Unassigned sessions are for anyone to grab. Our agent looks who worked on the related code last and pings people on the server. Whole thing was a prompt and team server extended itself since plugins are now hot reloadable."
View post on X ↗"I released an LLM plugin for sending prompts to OpenAI's new Jev-clone decision model that's based on GPT-6 Luna"
View post on X ↗"What are the top 1% of AI spenders paying so much money for? Tools to build, automate, and ship. Versus the average spender, they're: 23x more likely to pay for n8n, 18x for fal and Manus, 15x for Higgsfield. Full list of the Top 50 Consumer AI Apps by Revenue."
View post on X ↗"Great report from Google and MIT. Co-designing ML models and the accelerators is an exciting research direction. This report introduces Coco, an agent platform TPU architects use for hardware and model co-design. Each decision depends on hundreds of gigabytes of fresh data."
View post on X ↗