Here are today's news for 14th of October about OpenAI, Meta, Anthropic and Google DeepMind. Let's dive into the details. On October 7, 2026, OpenAI began rolling out GPT, 6 in ChatGPT with “Intelligent UI,” which can combine text with interactive charts, diagrams, forms, buttons and task, specific tools such as calculators, bill splitters and games. The global launch started with Plus, Pro, Business and Enterprise; Free and Go followed October 8. Paid plans use GPT, 6 Sol, while Free and Go use GPT, 6 Luna. OpenAI says the model selects a format suited to the prompt and can begin responding while it continues reasoning or using tools. The interface is built from streamable native components compiled as the response is generated, rather than being limited to static text or images. That shift could make ChatGPT more useful for learning, planning and calculations, while raising competitive pressure on rival assistants to deliver usable in, chat tools, not just answers. The feature is available from Instant through Extra High reasoning; no new price or benchmark figures were disclosed. On October 6, 2026, OpenAI published 722 manuscripts from an unreleased internal frontier model, arranged into 372 families of mathematical results. The company says the work includes solutions or substantial progress on hundreds of open problems across 17 fields, including number theory, geometry and theoretical computer science. The model was tested on about 4,000 problems; each reported result used computing equivalent, on average, to roughly three hours of ChatGPT Pro “thinking.” OpenAI also released reasoning summaries and Lean formalizations, which let computers check proofs. Among the highlighted claims are results concerning the Unique Games Conjecture and a special case of the Hodge conjecture; this is not a claim that the full Riemann hypothesis has been solved. The release could reshape mathematical research and intensify competition among AI labs, but independent scrutiny remains essential: the papers have not been peer, reviewed, and publication does not itself establish that every proof is correct. On October 12, 2026, Meta introduced Muse Wonder and Muse Wonder Full, calling them its most capable AI models to date and targeting coding and long, running agent tasks. Wonder became available through Meta’s Model API and Muse Code at $4 per million input tokens and $20 per million output tokens; the more capable Wonder Full is limited to Muse Code subscribers on its $50 and $200 monthly plans. Meta says the models can plan multi, step work, delegate to sub, agents, use tools in parallel and recover from errors; both also feature Active Memory and Active Focus. Meta reports Wonder Full leads independent DeepSWE and SWE, Marathon coding evaluations, though detailed comparisons are company, reported. The launch gives developers another contender against OpenAI and Anthropic, while Wonder is expected to power Meta’s Muse assistant, initially for subscribers. On Sept. 15, OpenAI policy chief Chris Lehane confirmed that OpenAI, Anthropic and Google DeepMind had held AI, safety discussions for several weeks. The talks have explored cooperation on safety standards and independent evaluations; no binding agreement, joint slowdown or formal standards body has been announced. The issue affects frontier, model developers, customers and regulators: shared testing could make safety claims more comparable, but cooperation among direct competitors raises questions about how common rules would work. Lehane said OpenAI did not believe an antitrust exemption was needed to discuss safety. The talks followed Anthropic CEO Dario Amodei’s Sept. 12 call to “pace the frontier” and give independent evaluators ongoing access to AI labs. OpenAI CEO Sam Altman backed external assessment, while Google DeepMind chair Demis Hassabis had previously proposed an independent standards body. OpenAI has also supported independent verification of frontier, AI safety practices. On September 30, Rep. Ro Khanna, the top Democrat on the House Select Committee on the Chinese Communist Party, sent letters to the CEOs of OpenAI, Anthropic, Google’s parent Alphabet, Meta and SpaceX’s AI unit. He requested records of known efforts by China or other hostile actors to obtain model weights or other intellectual property from the companies over the past two years, plus details of attempted intrusions, security incidents and defenses. The firms were asked to report cybersecurity spending and staffing, and whether independent experts had tested their safeguards; responses are due October 19. Model weights are the parameters that encode a system’s learned capabilities: stealing them could let an adversary reproduce a model while bypassing years of costly research. Khanna warned that theft could “erode America’s AI lead with the stroke of a keyboard. That's all for today's AI news briefing. Have a great day!