OpenAI ships GPT-5.5 and calls it the fastest path to a super app

OpenAI released GPT-5.5 on Thursday. Greg Brockman, OpenAI’s president and co-founder, called it the company’s “smartest and most intuitive” model yet. He also framed it as another step toward something the company has talked about for months: an AI super app.

What changed

GPT-5.5 is faster and cheaper per token than its predecessor 5.4. Brockman described it as “a faster, sharper thinker for fewer tokens.” The efficiency gains matter because they push frontier AI capabilities down into price tiers that more businesses and consumers can actually afford.

The benchmarks look strong on paper. OpenAI claims 5.5 outperforms Google’s Gemini 3.1 Pro and Anthropic’s Claude Opus 4.5 across most standard evaluations. Take those numbers with the usual grain of salt. Labs run their own comparisons and pick the tests that make them look best. But the directional claim, that 5.5 represents a meaningful jump over 5.4, tracks with what independent testers have been saying about OpenAI’s recent cadence.

Where it shines

Three areas got specific attention in the press briefing:

Agentic coding. The model handles multi-file code changes, debugging loops, and repository-level tasks better than prior versions. Mark Chen, OpenAI’s chief research officer, highlighted gains on “scientific and technical research workflows” too.

Computer use. Navigating interfaces, clicking through forms, completing multi-step tasks in browser environments. This is the capability that makes the super app vision plausible.

Research applications. Chen mentioned drug discovery explicitly. OpenAI thinks the model can help expert scientists make progress on actual hard problems, not just generate plausible-sounding text about them.

The super app angle

Brockman and Sam Altman have been hinting at this for months. The idea is to merge ChatGPT, Codex, and the AI browser into one unified product that enterprise customers can actually deploy as a platform. Not a chatbot that sometimes writes code. A single tool that does knowledge work end to end.

The timing is not accidental. Elon Musk wants X to become a super app. Every major AI lab is racing toward the same shape of product: a single interface between humans and all digital work. GPT-5.5 is OpenAI’s bet that model quality has crossed the threshold where that vision stops being marketing and starts being engineering.

Release cadence is accelerating

OpenAI shipped its previous model last month. Before that, December. Before that, November. Jakub Pachocki, OpenAI’s chief scientist, told reporters that “the last two years have been surprisingly slow.” He expects the pace to keep increasing.

Availability rolled out Thursday. Plus users get 5.5. Pro, Business, and Enterprise tiers get both 5.5 and 5.5 Pro. The Pro variant presumably gets more context window, higher rate limits, or both.

Competitive context

The release landed alongside a busy news day for AI. DeepSeek previewed its V4 model, claiming parity with American frontier systems. Google announced new TPUs built for what it calls the “agentic era.” Anthropic is dealing with fallout from unauthorized access to Mythos, its cybersecurity testing tool. Meta signed a deal for millions of Amazon AI CPUs while simultaneously laying off 10 percent of its workforce.