OpenAI's new plan REVEALED
TL;DR
OpenAI’s next model, codenamed “Spud,” has finished pre-training and could arrive within weeks — Wes says Sam Altman described it internally as a “very strong model” that could “really accelerate the economy,” which makes this sound bigger than a routine GPT update.
OpenAI is effectively killing Sora to free compute for a single all-in-one product push — according to Wes, the company is dropping side quests and consolidating ChatGPT, Codex, Operator/Atlas-style tools, and web browsing into a “super app” powered by Spud.
Altman is stepping back from direct safety oversight and focusing on chips, data centers, and supply chains — safety now reports to Mark Chen and security shifts under Greg Brockman, which Wes reads as a sign OpenAI is reorganizing around scaling infrastructure at “unprecedented” levels.
The AGI language inside OpenAI is getting more explicit — executive Fidji Simo reportedly renamed her org area to “AGI deployment,” the first time “AGI” has shown up on OpenAI’s org chart as a product-like category.
Sora’s shutdown may have blown up more than an internal roadmap — Wes claims a planned Disney licensing deal tied to AI video production, reportedly worth a $1 billion commitment over three years, is now dead before funds changed hands.
The biggest overlooked story is AI’s recent jump in math research credibility — Fields Medalist Terence Tao says Google DeepMind’s AlphaEvolve found the right approach to a proof and “ChatGPT proved the first and I proved the second,” which Wes treats as evidence that frontier AI is already becoming a real scientific co-worker.
The Breakdown
“Spud” lands, and Wes thinks this is the real story
Wes opens with OpenAI’s internal codename “Spud,” joking about the potato name but making clear he thinks this is a major release, not a minor refresh. Sam Altman told staff the model has finished pre-training, is “very strong,” and is coming in a few weeks — likely late this month or in April.
Altman reshuffles the org and moves closer to infrastructure
At the same time, Altman is stepping back from direct safety and security oversight: safety will report to Mark Chen, and security moves into scaling ops under Greg Brockman. Wes’s read is blunt: Altman wants bandwidth for the real bottleneck now — chips, data centers, power, and supply chains built at “unprecedented scale.”
OpenAI drops the side quests and goes all-in on a “super app”
Wes connects the dots by arguing OpenAI is done with fragmented products and is now building “the one app to rule all apps.” In his framing, ChatGPT, Codex, Atlas/Operator-style tooling, and web browsing are being folded into one product, and Spud is the model meant to power the whole thing.
Sora gets sacrificed, and the Disney angle makes it sting
This focus had casualties, and the biggest one is Sora: employees were reportedly shocked by how much compute it consumed, so OpenAI killed it to free resources for Spud and the super app. Wes says even planned ChatGPT-native video generation is off the roadmap, and he adds a dramatic detail — a Disney licensing deal tied to Sora, with a reported $1 billion production commitment over three years, appears to have died suddenly before money changed hands.
The Sora team isn’t disappearing — it’s pivoting to world models and robotics
Bill Peebles and the former Sora research team are now being redirected toward “systems that deeply understand the world by learning to simulate arbitrary environments at high fidelity.” Wes frames this as a pivot toward robotics world models, very much in line with what Google DeepMind and Nvidia have been signaling.
Anthropic is the pressure in the background
Wes says Altman’s memo explicitly names Google and Anthropic as the competition, and he emphasizes that Anthropic has been winning ground with enterprise users through products like Claude Code, Claude’s white-collar workflow tools, and Dispatch for managing agents from a phone. OpenAI’s recent acqui-hire of the OpenCloud creator, in his telling, looks like a direct attempt to catch up.
Then the video swerves to math — and this is where Wes gets animated
The “buried” companion story is Terence Tao, who recently published work showing how much AI has improved at theorem discovery. Wes zeroes in on Tao’s quote about a proof: AlphaEvolve found the right strategy, then “ChatGPT proved the first and I proved the second,” which he treats as a jaw-dropping signal that frontier models are no longer toy assistants.
Wes’s bottom line: stop betting against the curve
He argues the skeptics are increasingly cornered: if the world’s most respected living mathematician is publicly crediting AI for meaningful proof work, it’s getting harder to say the labs are faking progress. Wes ends by asking whether people still think claims about coding, science, and economic acceleration were just hype — because from his view, the people “following the line on the chart” have been more right than the people dismissing it.