The Omniscient Bulletin · 2026-10-07
The Omniscient Bulletin — October 7, 2026
OpenAI publishes 722 math manuscripts as Mistral previews a 1 trillion parameter model
OpenAI's math catalogue arrives with its own warning that some unformalized results could have issues, and Mistral's largest model is a preview whose weights are still to come. Google is paying to add nuclear capacity at existing Constellation plants, and the Justice Department told staff to call AI super intelligence, even in court filings when appropriate.
Frontier
OpenAI published 722 math manuscripts from an unreleased internal model, and says the collection holds results at different stages of verification
OpenAI released a GitHub catalogue of 722 manuscripts, organized into 372 families, produced by an unreleased internal model. The model was posed approximately 4,000 problems, and each result took, on average, three hours of ChatGPT Pro thinking compute. OpenAI says the results in the collection are at different stages of verification, that not all have accompanying Lean formalizations, and that "some of the unformalized results could have issues." It says it will record corrections and revisions as new versions.
Frontier
Mistral opened a public preview of Large 4, a 1 trillion parameter model, and says it will release the weights by the end of the month
Mistral launched a public preview of Mistral Large 4, a natively multimodal model with 1 trillion parameters, 49 billion of them active, which it calls its largest and most capable to date. Mistral trained it from scratch, using 3,800 Nvidia Grace Blackwell GPUs, at its own European data centers. The weights are not out yet; Mistral says it will release them by the end of the month, and until then cybersecurity leaders, vetted partners and state authorities are red-teaming a version with lighter moderation and broader cyber capabilities.
Frontier
Anthropic merged Project Glasswing into a three-tier cyber access program, and currently reviews every organization in its Specialized Access tier with the US government
Anthropic launched an expanded Cyber Verification Program with three access tiers, giving qualifying security teams Claude Opus 5.5, Sonnet 5.5 and Mythos 5.1 with reduced blocking classifiers. Anthropic currently vets every Specialized Access organization in depth with the US government, and Glasswing members move into that tier. It says partners found at least 129,000 verified vulnerabilities from April to July, a count it calls likely an undercount, and its own open-source scanning found 5,500 more.
Compute
Google signed a 20-year deal funding 890 MW of nuclear uprates at Constellation plants, plus a 15-year supply agreement for 2,700 MW more
Google and Constellation announced a 20-year power purchase agreement that funds upgrades at 11 Constellation nuclear units across Illinois, Pennsylvania and New Jersey, bringing 890 MW of new capacity onto the PJM grid, the first expected by 2028. Constellation puts its investment at more than $4.3 billion. The companies also signed a 15-year energy supply deal for another 2,700 MW in PJM, and Constellation chose Google Cloud and Gemini Enterprise in a broader five-year technology alliance.
Compute
SpaceX is seeking $40 billion in financing led by Apollo to buy Nvidia AI chips, the Financial Times reported
SpaceX is seeking about $40 billion, $10 billion in bank loans and $30 billion in investment-grade debt, to buy Nvidia AI chips, the Financial Times reported, citing people familiar with the matter. Apollo is expected to lead the deal and help place the debt, with Pimco among a few lenders in talks, and the transaction is expected to close in 2027. Musk has said SpaceX will build its data centers only with Nvidia hardware. SpaceX, Apollo and Nvidia did not immediately respond to Reuters requests for comment.
Policy
The Justice Department told staff to say super intelligence instead of artificial intelligence, including in court filings when appropriate
The Justice Department on Tuesday told its employees to use super intelligence, or SI, in place of artificial intelligence, according to a memo seen by Reuters. Acting Deputy Attorney General Trent McCotter's directive applies "to the maximum extent permitted by law" in public communications, policy documents and other official records, and "shall extend to court filings when appropriate." It follows an executive order Trump issued last week requiring federal agencies to make the switch.
Labor
HubSpot is cutting about 7% of its team, nearly 660 people, and told staff the cut is not driven by AI efficiencies
HubSpot is cutting about 7% of its workforce, nearly 660 people, according to a staff memo filed with the SEC. CEO Yamini Rangan wrote that the cut is "not driven by AI-related efficiencies," and that the company has shifted "from building software that helps customers grow to delivering outcomes for them with AI." The same day, FICO disclosed it will eliminate about 15% of positions, citing in part the integration of AI-driven product development.