OpenAI’s GPT-5.2 Escalates the AI Agent War
▼ Summary
– OpenAI has released GPT-5.2, which it describes as its best model yet for professional use, with improvements in tasks like creating spreadsheets, writing code, and handling complex projects.
– The model is presented as a competitor to Google’s Gemini 3 and is noted for generating sharper, more explanatory responses in expert fields like immunology.
– OpenAI highlights that the new “Thinking” model hallucinates less than its predecessor, making it more reliable for professional and agentic AI workflows.
– The release follows internal pressure, with CEO Sam Altman declaring a “code red” to prioritize ChatGPT improvements over initiatives like advertising.
– The launch coincides with a major Disney licensing deal and the beginning of a rollout for an age-prediction model intended to apply safeguards for minors.
OpenAI has launched GPT-5.2, positioning it as its premier model for professional tasks and intensifying the competition with rivals like Google’s Gemini. The company asserts this new series, which includes Instant, Thinking, and Pro variants, represents a significant leap in capability for handling complex workflows, from data analysis and coding to long-form reasoning and creative projects. This release is a direct challenge in the escalating AI agent war, where companies vie to offer the most efficient and useful automated assistants.
During a briefing, OpenAI executives emphasized the model’s practical value. Fidji Simo, CEO of Applications, stated the development took “many, many months” with the goal to “unlock even more economic value for people.” The company highlighted specific professional strengths, noting GPT-5.2 excels at creating spreadsheets, building presentations, writing code, perceiving images, and managing multi-step projects. In a test case, a senior immunology researcher reportedly found that the Pro model generated “sharper questions and stronger explanations” on complex topics than other leading systems.
A key focus is enhancing AI agents, software that can autonomously perform tasks. Researcher Max Schwarzer outlined a vision for ChatGPT to become the ultimate “personalized assistant.” The Thinking model is specifically noted to “hallucinate less” than its predecessor, a critical improvement for professionals who require trustworthy, agentic tools. Major companies like Notion, Box, Shopify, and Zoom were among the pre-release testers granted access weeks ago.
This update follows GPT-5.1 by less than a month. While that earlier version focused on making interactions “warmer” and introducing personality presets, GPT-5.2 builds upon that foundation. OpenAI promises it will feel “more structured, more reliable, and still enjoyable to talk to” for daily use.
The aggressive release schedule occurs against a backdrop of intense rivalry. OpenAI is reportedly responding to Google’s advances and has internally declared a “code red,” reallocating resources to improve ChatGPT core capabilities, even delaying initiatives like advertising. Simo confirmed the internal shift was to “marshal resources in this one particular area,” though she declined to provide new details on advertising timelines.
In a related major business development, OpenAI announced a three-year licensing deal with Disney on the same day. The partnership allows for user-prompted social videos featuring over 200 characters from Disney, Marvel, Pixar, and Star Wars. Disney will stream some videos on Disney Plus and becomes a “major customer” alongside making a substantial equity investment.
OpenAI also provided updates on safety features, mentioning an “early stages” rollout of an age-prediction model designed to auto-apply safeguards for minors. Simo said testing is beginning in select countries to accurately identify teens without misidentifying adults. This system is a precursor to a planned “adult mode” for ChatGPT, tentatively slated for early 2026.
The GPT-5.2 models are rolling out starting today for paid ChatGPT plans (Plus, Pro, Go, Business, and Enterprise), with a gradual deployment to ensure a smooth experience. The previous GPT-5.1 model will remain accessible under a “legacy models” menu for three months before being retired.
(Source: The Verge)
