Everyone needs a Muse

The Ben's Bites newsletter for September 10, 2026 leads with several product and research items. OpenAI released ChatGPT Images 2.5, described as a new image generation model that is extremely good at editing any element in an image while keeping everything else the same, and taking up to 50% less time per image. ChatGPT also added Sketch, a feature that turns scribbles into a generated picture.

Meta introduced Muse, described as a personal agent and Zuckerberg's competitor to Grok Bot, Instinct and others. The newsletter says it offers the same functionality, including booking travel, handling email and working on any task on a computer, with Meta arguing it is built much more securely and can connect Facebook and Instagram accounts. It is initially for adults in the US only.

The newsletter states OpenAI solved the Navier-Stokes problem, one of the seven Millennium Problems given that status in 2000, and says the process was dramatic, involving authorship disputes. It attributes the effort to OpenAI acting after hearing rumours that Anthropic had solved a Millennium Prize problem too.

Separately, Jacob Coxon, an Anthropic and ex-OpenAI employee, resigned on Twitter citing safety concerns, in what the newsletter calls likely the most-viewed AI safety comms ever, while noting many are questioning whether it is a conspiracy. Another safety figure is joining the OpenAI board, and an alignment report on Claude was published.

Other items include a movie about Sam Altman's firing and rehiring, titled Artificial, releasing this Christmas with Andrew Garfield as Altman; an OpenAI essay on the window for AI policy; and in the author's notes, work using Astra and Fable 5.1 to orchestrate other agents and experimentation with open-source models such as GLM 5.3.