OpenAI Science Division Lasted 7 Months Before Codex Won

#openai #chatgpt #codex #wired

The OpenAI science division lasted about seven months as a named initiative. Kevin Weil announced OpenAI for Science in September 2025. Prism, its scientist-facing web app, launched in January 2026. By April, WIRED reported that Weil was leaving, Prism was being sunset, and the roughly 10-person Prism team was being folded under Codex.

That is a faster reversal than the headlines suggest. The obvious read is executive churn. The better read is organizational: OpenAI appears to have decided that scientific tooling does not get to stay standalone unless it strengthens the main product stack quickly.

I started out thinking this was mostly about Kevin Weil leaving OpenAI. The reporting points somewhere more interesting. OpenAI is collapsing a fresh science initiative into its coding product at the same time it says it wants to “unify its business and product strategy.” In plain English: if a tool can help make Codex into an “everything app,” it lives. If not, it gets absorbed.

Why the OpenAI science division is folding into Codex

The confirmed facts are straightforward. WIRED reports that OpenAI is sunsetting Prism, the web app it launched in January to help scientists work with AI. WIRED also reports that OpenAI is moving the roughly 10-person Prism team under Thibault Sottiaux, OpenAI’s head of Codex, with plans to bring Prism’s capabilities into the desktop Codex app. An OpenAI spokesperson confirmed that this is part of an effort to unify business and product strategy.

That is verified. The motive beyond that is partly interpretation, but the pattern is hard to miss.

OpenAI has already been narrowing its product surface. WIRED says Fidji Simo told staff in March that the company needed to simplify its offerings, and that this push contributed to shutting down the Sora app. We covered that in OpenAI Sora Shutdown. Now the same logic appears to be hitting science tooling.

The strange part is the timing. Weil announced OpenAI for Science in September 2025. Prism shipped in January 2026. WIRED’s reporting on OpenAI’s coding push still described Weil as leading OpenAI for Science just weeks ago, with the ambition to make 2026 “for science what 2025 was for software engineering.” That is not a long runway. By big-company standards, Prism barely made it out of onboarding.

Initiative	Launch / Role	What was promised	What happened
OpenAI for Science	Announced Sept. 2025	A dedicated science initiative	Verified: decentralized into other teams
Prism	Launched Jan. 2026	Better AI workspace for scientists	Verified: sunset; capabilities planned for Codex
Codex	Existing coding app	Coding assistant, now broader platform	Verified: OpenAI wants it to become an “everything app”

The cleanest explanation is that Codex won the internal resource fight. Not because science stopped mattering, but because science had to justify itself as a product.

What Kevin Weil’s exit signals about OpenAI’s priorities

We know Kevin Weil leaving OpenAI is real. WIRED confirmed his departure, and Weil posted that “Today is my last day at OpenAI, as OpenAI for Science is being decentralized into other research teams.” That part is not rumor.

What we do not know is the exact direction of causality. Did Weil leave because the science initiative was being dissolved? Or did the initiative get dissolved because Weil was leaving? The current reporting does not establish that. Treat any confident answer here as unverified.

Still, the surrounding evidence points to a company prioritizing a smaller number of commercial lanes. WIRED says OpenAI is refocusing around enterprise offerings and coding as it faces pressure from Anthropic and prepares to file for an IPO later this year. TechCrunch describes the broader move as shedding “side quests.” That phrasing is theirs, but the examples line up: Sora is gone, Prism is being folded in, and Codex keeps getting promoted.

That tracks with OpenAI’s recent product behavior. Coding is measurable, sticky, and monetizable. Enterprise buyers understand it. Benchmarks help sell it. Scientists are a real market, but a much less legible one inside a company trying to simplify, grow revenue, and win the developer workflow. If you want the less romantic version: one seat of Codex is easier to price than “accelerating discovery.”

There is also a personnel signal here. Weil moved from chief product officer into a science role, then exits as the standalone effort disappears. That does not prove failure of the science idea. It does suggest that, inside OpenAI, “science” did not become important enough to remain its own power center.

Prism’s shutdown shows the product-first trade-off

Prism is the most concrete piece of evidence because it was an actual shipped product. OpenAI launched it in January as a web app for scientists. By April, it was being sunset. That is verified by WIRED.

The company says Prism’s capabilities will be incorporated into Codex. That is a plausible plan, not yet a delivered outcome. Readers should keep those separate. Shipping a standalone scientist workflow is different from preserving those features after they are moved into a broader desktop app with many other priorities. Product roadmaps are full of promised integrations that become menu items and then become memories.

The trade-off is easy to state and hard to avoid:

A standalone science app can optimize for research workflows.
A unified Codex app can reuse distribution, identity, billing, and model interfaces.
Companies under pressure usually pick the second one.

OpenAI is not unusual here. It is just unusually visible. Frontier labs increasingly look like software companies with expensive research departments attached. That means internal projects are judged less by whether they are admirable and more by whether they compound the core platform.

That also helps explain why coding keeps winning. Coding products already sit near OpenAI’s center of gravity: model evals, enterprise adoption, developer mindshare, and now the broader “AI builds AI” loop. We wrote about that dynamic in AI Builds AI. A science product may matter strategically, but a coding product improves the machine that builds the next coding product. Executives tend to notice that.

What the OpenAI science division reset means for scientists and builders

For scientists, the immediate implication is boring and inconvenient. Prism users now have a sunset product and a promise. Maybe the useful parts reappear inside Codex. Maybe they return in a form optimized for a much broader audience. Maybe some of the sharper science-specific edges get sanded off in the merge. Right now, only the shutdown is confirmed.

For builders, the lesson is clearer. Watch what gets merged into the company’s main app. That tells you more than the launch blog posts.

OpenAI can still credibly say it cares about scientific discovery. WIRED notes the company announced GPT-Rosalind models for life sciences researchers the same day. That is verified. But the organization chart is making a different point: science is welcome as a capability layer, not necessarily as a standalone product surface.

That matters if you are building on top of OpenAI. The safest bets are the ones that align with the company’s current spine: enterprise, coding, and consolidated desktop workflows. If your use case sits outside that spine, assume you are renting from a moving landlord.

It also matters for the bigger OpenAI narrative. The company is still growing aggressively — see our breakdown of OpenAI revenue 2026 — but growth usually comes with simplification, not expansion in every direction. The OpenAI science division story is what that looks like internally. Not “science is over.” More like: science has to justify itself in Codex-shaped terms now.

Key Takeaways

Verified: Kevin Weil is leaving OpenAI, OpenAI for Science is being decentralized, and Prism is being sunset.
Verified: Prism’s roughly 10-person team is moving under Codex, with plans to bring Prism capabilities into the Codex app.
Unverified: The exact causal link between Weil’s exit and the science reorganization is still unclear.
The real signal: OpenAI appears to be consolidating around coding, enterprise, and fewer flagship products.
For builders: Watch the core app, not the side initiative. That is where OpenAI is placing its durable bets.