Author: Rebecca Bellan

OpenAI launched its latest frontier model, GPT-5.2, on Thursday amid increasing competition from Google, pitching it as its most advanced model yet and one designed for developers and everyday professional use. 

OpenAI’s GPT-5.2 is coming to ChatGPT paid users and developers via the API in three flavors: Instant, a speed-optimized model for routine queries like information-seeking, writing, and translation; Thinking, which excels at complex structured work like coding, analyzing long documents, math, and planning; and Pro, the top-end model aimed at delivering maximum accuracy and reliability for difficult problems. 

“We designed 5.2 to unlock even more economic value for people,” Fidji Simo, OpenAI’s chief product officer, said Thursday during a briefing with journalists. “It’s better at creating spreadsheets, building presentations, writing code, perceiving images, understanding long context, using tools and then linking complex, multi-step projects.”

GPT-5.2 lands in the middle of an arms race with Google’s Gemini 3, which is topping LMArena’s leaderboard across most benchmarks (apart from coding — which Anthropic’s Claude Opus-4.5 still has on lock).

Early this month, The Information reported that CEO Sam Altman released an internal “code red” memo to staff amid ChatGPT traffic decline and concerns that it is losing consumer market share to Google. The code red called for a shift in priorities, including stalling on commitments like introducing ads and instead focusing on creating a better ChatGPT experience. 

GPT-5.2 is OpenAI’s push to reclaim leadership, even as some employees reportedly asked for the model release to be pushed back so the company could have more time to improve it. And despite indications that OpenAI would focus its attention on consumer use cases by adding more personalization and customization to ChatGPT, the launch of GPT-5.2 looks to beef up its enterprise opportunities. 

The company is specifically targeting developers and the tooling ecosystem, aiming to become the default foundation for building AI-powered applications. Earlier this week, OpenAI released new data showing enterprise usage of its AI tools has surged dramatically over the past year. 

This comes as Gemini 3 has become tightly integrated into Google’s product and cloud ecosystem for multimodal and agentic workflows. Google this week launched managed MCP servers that make its Google and Cloud services like Maps and BigQuery easier for agents to plug into. (MCPs are the connectors between AI systems and data and tools.)

OpenAI says GPT-5.2 sets new benchmark scores in coding, math, science, vision, long-context reasoning, and tool use, which the company claims could lead to “more reliable agentic workflows, production-grade code, and complex systems that operate across large contexts and real-world data.”

Those capabilities put it in direct competition with Gemini 3’s Deep Think mode, which has been touted as a major reasoning advancement targeting math, logic, and science. On OpenAI’s own benchmark chart, GPT-5.2 Thinking edges out Gemini 3 and Anthropic’s Claude Opus 4.5 in nearly every listed reasoning test, from real-world software engineering tasks (SWE-Bench Pro) and doctoral-level science knowledge (GPQA Diamond) to abstract reasoning and pattern discovery (ARC-AGI suites). 

Research lead Aidan Clark said that stronger math scores aren’t just about solving equations. Mathematical reasoning, he explained, is a proxy for whether a model can follow multi-step logic, keep numbers consistent over time, and avoid subtle errors that could compound over time. 

“These are all properties that really matter across a wide range of different workloads,” Clark said. “Things like financial modeling, forecasting, doing an analysis of data.”

During the briefing, OpenAI product lead Max Schwarzer said GPT-5.2 “makes substantial improvements to code generation and debugging” and can walk through complex math and logic step by step. Coding startups like Windsurf and CharlieCode, he added, report “state-of-the-art agent coding performance” and measurable gains on complex multi-step workflows.

Beyond coding, Schwarzer said that GPT-5.2 Thinking responses contain 38% fewer errors than its predecessor, making the model more dependable for day-to-day decision-making, research, and writing. 

GPT-5.2 appears to be less a reinvention and more of a consolidation of OpenAI’s last two upgrades. GPT-5, which dropped in August, was a reset that laid the groundwork for a unified system with a router to toggle the model between a fast default model and a deeper “Thinking” mode. November’s GPT-5.1 focused on making that system warmer, more conversational, and better suited to agentic and coding tasks. The latest model, GPT-5.2, seems to turn up the dial on all of those advancements, making it a more reliable foundation for production use. 

For OpenAI, the stakes have never been higher. The company has made commitments to the tune of $1.4 trillion for AI infrastructure buildouts over the next few years to support its growth — commitments it made when it still had the first-mover advantage among AI companies. But now that Google, which lagged behind at the start, is pushing ahead, that bet might be what’s driving Altman’s “code red.” 

OpenAI’s renewed focus on reasoning models is also a risky flex. The systems behind its Thinking and Deep Research modes are more expensive to run than standard chatbots because they chew through more compute. By doubling down on that kind of model with GPT-5.2, OpenAI may be setting up a vicious cycle: spend more on compute to win the leaderboard, then spend even more to keep those high-cost models running at scale.

OpenAI is already reportedly spending more on compute than it had previously let on. As TechCrunch reported recently, most of OpenAI’s inference spend — the money it spends on compute to run a trained AI model — is being paid in cash rather than through cloud credits, suggesting the company’s compute costs have grown beyond what partnerships and credits can subsidize.

During the call, Simo suggested that as OpenAI scales, it is able to offer more products and services to generate more revenue to pay for additional compute.

“But I think it’s important to place that in the grand arc of efficiency,” Simo said. “You are getting, today, a lot more intelligence for the same amount of compute and the same amount of dollars as you were a year ago.”

For all its focus on reasoning, one thing that’s absent from today’s launch is a new image generator. Altman reportedly said in his code red memo that image generation would be a key priority moving forward, particularly after Google’s Nano Banana (the nickname for Google’s Gemini 2.5 Flash Image model) had a viral moment following its August release.

Last month, Google launched Nano Banana Pro (aka Gemini 3 Pro Image), an upgraded version with even better text rendering, world knowledge, and an eerie, real-life, unedited vibe to its photos. It also integrates better across Google’s products, as demonstrated over the past week as it pops up in tools and workflows like Google Labs Mixboard for automated presentation generation.

OpenAI reportedly plans to release another new model in January with better images, improved speed, and better personality, though the company didn’t confirm these plans Thursday.

OpenAI also said Thursday it’s rolling out new safety measures around mental health use and age verification for teens, but didn’t spend much of the launch pitching those changes.

This article has been updated with more information about OpenAI’s compute efficiency status.

Got a sensitive tip or confidential documents? We’re reporting on the inner workings of the AI industry — from the companies shaping its future to the people impacted by their decisions. Reach out to Rebecca Bellan at rebecca.bellan@techcrunch.com or Russell Brandom at russell.brandom@techcrunch.com. For secure communication, you can contact them via Signal at @rebeccabellan.491 and russellbrandom.49.

Ref link: OpenAI fires back at Google with GPT-5.2 after ‘code red’ memo