AI news, October 28, 2026: Genie 4, Opera's local WebAI, OpenAI's ETH claim
Five stories from Monday and Tuesday, October 26โ27, 2026 (US time) for solo makers who build with AI. Two are things you can try or plan around soon, one is a big research claim that still needs checking, and two touch on how agents and models get run. Figures are as the companies and outlets state them.
Google DeepMind releases Genie 4, a world model with sound and mid-run prompts
Google DeepMind put Genie 4 out as a research preview for trusted testers. It generates interactive, more photorealistic worlds, lets you type new prompts while the simulation runs to change the scene, and produces matching real-time audio such as music, ambient sound and dialogue. The team told TechCrunch that objects can still drift in long runs and in-world text is not reliably legible. It is due to reach Google AI Ultra subscribers in Project Genie and early-access Gemini API and AI Studio partners in the coming weeks.
Why it matters for makers: generated, steerable worlds could become a fast way to prototype game levels or scenes, but access is limited for now.
Source: TechCrunch
Opera ships WebAI, letting websites use AI models that run on your own device
Opera introduced WebAI as an opt-in early preview in Opera One, with Opera GX and Opera Neon to follow. It lets websites and extensions call locally running open-weight models, such as Google's Gemma, and Apple's on-device model on Macs, so prompts never leave the computer. Opera pitches it as a way for developers to add AI features without running servers or paying per request; the preview covers 64-bit Windows PCs and Apple-silicon Macs.
Why it matters for makers: if browser-local AI catches on, small web apps could offer AI features at zero inference cost, though for now it reaches only Opera users who switch it on.
Sources: Opera blog, Opera Newsroom, TechCrunch
OpenAI says its models disproved the Exponential Time Hypothesis
OpenAI announced three results from its internal models: a claimed disproof of the Exponential Time Hypothesis, a long-standing assumption about how fast problems like SAT can be solved; a proof of the local smoothing conjecture in every dimension; and results in quantum field theory. The computer science proof and its Lean formalization are public on GitHub, while the physics group is writing up its work before releasing the AI output. OpenAI calls the manuscripts "raw research data" and has asked outside experts to lead the review instead of claiming academic credit.
Why it matters for makers: none of this is peer-reviewed yet, so treat it as a claim to watch, not a settled result.
Sources: OpenAI, The Verge, Al Jazeera
GitHub Copilot lets apps and agents authenticate as an organization
GitHub added server-to-server authentication for the Copilot SDK in public preview. A GitHub App with the read-only "Copilot Requests" organization permission can pass a short-lived installation token, so a service or agent calls Copilot on the organization's behalf without a signed-in user. Usage is billed to the organization, its model policies and budgets apply, and owners can cut off an app by suspending or uninstalling it.
Why it matters for makers: bots that run on a schedule no longer have to borrow one person's Copilot login.
Source: GitHub Changelog
Nvidia backs Simplismart, which helps companies run models on fewer GPUs
Indian startup Simplismart raised a $14 million Series B led by Dallas Venture Capital, with Nvidia, Accel, Shastra VC and Micromax taking part. The company makes software that runs AI models more efficiently, in some cases on fewer GPUs, as demand for compute keeps outrunning supply.
Source: TechCrunch