AI news, 7 November: OpenAI locks in $11 billion at $1.25 trillion valuation

OpenAI closes major funding tranche. OpenAI has completed the first portion of a new funding round, securing roughly $11 billion at a $1.25 trillion valuation, according to Bloomberg and Reuters citing people familiar with the matter. The funds were wired as of Friday. The full round could reach about $20 billion in coming weeks, well below the company’s earlier $30 billion target at a higher valuation, as public-market sentiment toward AI cooled. Investors in the initial close reportedly include SoftBank, Coatue, T. Rowe Price and others. OpenAI declined to comment. The raise comes amid broader market pressure and the company’s recent cybersecurity model work.

Independent testers say Anthropic’s progress has roughly doubled. Nonprofit evaluator METR published an assessment finding that Anthropic’s models have been improving capabilities on well-specified tasks about 1.9 times faster since October 2025 than in the prior period, packing roughly two years of earlier progress into one. METR estimates that around 80 percent of the speedup comes from AI assisting the company’s own research and development rather than extra compute, based on internal data and staff interviews during an 11-week review. The group noted high uncertainty and that Anthropic redacted key evidence; the finding adds weight to discussions about AI accelerating its own advance.

AI system helps close long-open quantum conjecture. Research organization PSI reported that an internal system running non-public Anthropic model instances produced a complete proof of the Aaronson-Ambainis conjecture, open since 2008, by the third day of a run. The Lean proof assistant accepted it, and the system later improved the result. The conjecture concerns limits on quantum speedups for certain random problems. Verification so far rests on the formal Lean check plus an internal advisor audit; there is no peer-reviewed journal publication yet. Human researchers scoped the problem, reviewed the draft and take responsibility for correctness.

Google lengthens and cheapens AI video generation. Google made Gemini Omni 1.2 Flash generally available, supporting video clips up to 15 seconds (up from 10) with multi-turn conversational editing, reference images and audio. Pricing falls about 30 percent to $0.07 per second for 720p, according to the company. It is offered via the Gemini API, AI Studio, Cloud and the app for higher-tier users, increasing pressure on rival video tools.

Major labs back shared cyber-jailbreak rating system. The Frontier Model Forum, whose members include OpenAI, Anthropic, Google DeepMind, Microsoft and Amazon, released Version 1.0 of a framework for rating the risk of successful cyber jailbreaks. It scores four dimensions (capability gain, scope, ease of deployment and discovery) into risk bands from Negligible to Critical. The preliminary tool aims to help developers, researchers and policymakers compare findings consistently; public comments are open until 21 December.

Markets react to data-centre politics. Tech shares sold off sharply after US midterms, with the Nasdaq 100 losing around $2.3–2.5 trillion in value in one session as investors fretted that Democratic gains and local ballot measures could slow AI data-centre construction. Separate coverage noted OpenAI pledging to cover certain grid and infrastructure costs and end NDAs with public officials on future campuses. These remain market and political developments rather than final policy.