OpenAI pivots again: doubling down on text models, shelving video, and building a “superapp”

The pivot

OpenAI President Greg Brockman told Big Technology that the company is refocusing its efforts around large language models — and that means fewer experiments with video and more investment in text-driven assistants. It has been reported that OpenAI shut down its video-generation project (referred to as Sora) as part of this shift. Brockman said the team now has “line of sight to much better models coming this year,” and that the company believes text models are the right architecture to push toward AGI.

Why text, why now?

Brockman framed the move as pragmatic. After years of cerebral demos and benchmark work, OpenAI wants real-world feedback from knowledge workers and business users. What matters, he argued, are two priorities: a personal assistant that knows you, and AI that can solve hard problems with full context. Can one team do everything? Not with current compute and priorities. So choices had to be made.

Codex, the superapp and compute

The interview also dove into Codex’s potential and the plan to stitch coding, chat, and browsing into a single “superapp” — think an ambitious, productivity-first Swiss Army knife rather than a consumer gimmick. Brockman described a vision where AI uses your computer to make you more effective at work and assists “everywhere else.” He defended the heavy compute spend as necessary to realize those ambitions, arguing that large language models are surprisingly extensible into domains people once thought required dedicated world or video models.

The gamble and the questions left hanging

This is a strategic bet: double down on text, consolidate features into a superapp, and prioritize business impact. It is a big swing. Will narrowing the roadmap and shuttering projects like Sora accelerate real-world usefulness — or leave money on the table in creative and consumer spaces? Brockman sounds confident. Time, and a few model releases, will tell.

Sources: bigtechnology.com