OpenAI releases GPT-5.2

OpenAI released GPT-5.2, an update to its GPT-5 line, in three configurations — Instant, Thinking and Pro — available immediately to paid ChatGPT users and developers through the API. OpenAI positioned it around professional knowledge work rather than a single headline capability, citing improvements in spreadsheet and presentation creation, image understanding, long-context handling and coding, alongside gains on science and mathematics benchmarks.

The company’s central claimed metric was GDPval, an internal-style benchmark spanning 44 occupations that compares model output to that of industry professionals: OpenAI reported GPT-5.2 Thinking tying or beating professionals on 70.9% of comparisons, against 38.8% for GPT-5.1. It also reported gains on coding benchmarks including SWE-bench Verified and SWE-bench Pro, and said hallucination rates on real ChatGPT queries had fallen relative to GPT-5.1.

The release came roughly three weeks after Google shipped Gemini 3, which had taken a clear lead on independent leaderboards and benchmark comparisons and, according to later reporting, prompted an internal “code red” at OpenAI — with chief executive Sam Altman said to have told staff to deprioritise other initiatives, including planned advertising features, in favour of improving ChatGPT’s core product. Speaking to CNBC on the day of the release, Altman said Gemini 3’s impact on ChatGPT usage had been smaller than initially feared, and that he expected OpenAI to exit “code red” status by January.

As with prior releases in the sequence, GPT-5.2’s benchmark figures were self-reported by OpenAI rather than independently verified, and the rapid, competitively-timed cadence of GPT-5, GPT-5.1 and GPT-5.2 releases within months of each other was itself read by industry observers as evidence of how directly Google’s Gemini 3 had disrupted OpenAI’s product roadmap.