OpenAI launched its latest Frontier model, GPT-5.2, on Thursday as competition from Google intensifies, touting it as its most advanced model yet and one designed for developers and everyday professionals.
OpenAI’s GPT-5.2 is available via API to ChatGPT paid users and developers in three flavors: Instant is a speed-optimized model for everyday queries such as information search, writing, and translation. Thinking ability. Excels at complex structured tasks such as coding, analyzing long documents, mathematics, and planning. Pro is a top-of-the-line model designed to provide maximum accuracy and reliability for difficult problems.
“We designed 5.2 to unlock even more economic value for people,” Fiji Simo, OpenAI’s chief product officer, said in a press conference with journalists on Thursday. “I’m good at creating spreadsheets, building presentations, writing code, recognizing images, understanding long contexts, using tools, and linking complex multi-step projects.”
GPT-5.2 is in an arms race with Google’s Gemini 3. Gemini 3 tops LMArena’s leaderboard in most benchmarks (coding aside, Anthropic’s Claude Opus-4.5 is still rocking).
Earlier this month, The Information reported that CEO Sam Altman released an internal “Code Red” memo to staff amid concerns about declining ChatGPT traffic and losing consumer market share to Google. Code Red required a change in priorities, including stalling efforts like introducing ads and instead focusing on creating a better ChatGPT experience.
GPT-5.2 is an effort by OpenAI to regain control, with some employees reportedly asking for the model release to be delayed so the company could spend more time making improvements. And despite indications that OpenAI will focus on consumer use cases by adding more personalization and customization to ChatGPT, the release of GPT-5.2 looks to strengthen opportunities in the enterprise.
The company is specifically targeting developers and the tools ecosystem, aiming to become the default foundation for building AI-powered applications. Earlier this week, OpenAI released new data showing that enterprise usage of its AI tools has increased dramatically over the past year.
tech crunch event
san francisco
|
October 13-15, 2026
This comes as Gemini 3 is tightly integrated into Google’s product and cloud ecosystem for multimodal and agent workflows. Google this week launched a managed MCP server that makes it easy for agents to connect its Google services with cloud services like Maps and BigQuery. (MCP is a connector between AI systems and data and tools.)
According to OpenAI, GPT-5.2 sets new benchmark scores in coding, math, science, vision, long-context reasoning, and tool usage, which the company claims can lead to “more reliable agent workflows, production-grade code, and complex systems that operate across large scale contexts and real-world data.”
These features compete directly with Gemini 3’s Deep Think mode, which is touted as a major advancement in reasoning for math, logic, and science. In OpenAI’s own benchmark charts, GPT-5.2 Thinking outperforms Gemini 3 and Anthropic’s Claude Opus 4.5 in nearly every reasoning test listed, from real-world software engineering tasks (SWE-Bench Pro) and PhD-level scientific knowledge (GPQA Diamond) to abstract reasoning and pattern discovery (ARC-AGI suite).
Lead researcher Aidan Clarke said improving maths performance was not just about solving equations. He explained that mathematical reasoning is a way to determine whether a model can follow multi-step logic, maintain numerical consistency over time, and avoid subtle errors that can worsen over time.
“These are all really important characteristics across a variety of workloads,” Clark said. “Financial modeling, forecasting, data analysis, etc.”
OpenAI product lead Max Schwarzer said in a briefing that GPT-5.2 “brings significant improvements to code generation and debugging,” allowing complex math and logic to be executed step-by-step. Coding startups like Windsurf and CharlieCode are reporting “state-of-the-art agent coding performance” and measurable benefits in complex multi-step workflows, he added.
Beyond coding, Schwarzer said GPT-5.2 Thinking responses have 38% fewer errors than their predecessors, making the model more reliable in everyday decision-making, research, and writing.
GPT-5.2 seems less like a reinvention and more like an amalgamation of OpenAI’s previous two upgrades. GPT-5, released in August, was a reset that laid the foundation for an integrated system with routers that switch models between a fast default model and a deeper “thinking” mode. GPT-5.1 in November focused on making the system warmer, more conversational, and better suited for agent and coding tasks. The latest model, GPT-5.2, further enhances all these advances and appears to be a more reliable foundation for use in production environments.
For OpenAI, the stakes have never been higher. The company has committed to investing $1.4 trillion in building AI infrastructure over the next few years to support growth. This is a promise made when there was still first-mover advantage among AI companies. But Google, which initially lagged behind, is now moving forward, and that bet may be driving Altman’s “Code Red.”
OpenAI’s new focus on inference models is also a risky change. The systems behind Thinking and Deep Research modes perform more computing and are therefore more expensive to run than a standard chatbot. By doubling down on that kind of model with GPT-5.2, OpenAI may be creating a vicious cycle. That means spending more on compute to get to the leaderboards, and then spending even more to keep these expensive models running at scale.
OpenAI is already reportedly spending more on computing than before. As TechCrunch recently reported, most of OpenAI’s inference spending (money spent on computing to run trained AI models) is paid for in cash rather than cloud credits, suggesting that the company’s computing costs are growing beyond what can be subsidized with partnerships or credits.
During the conference call, Simo suggested that as OpenAI scales, it could offer more products and services and increase the revenue to pay for additional compute.
“But I think it’s important to place it within the larger arc of efficiency,” Simo said. “Today, you can get more intelligence for the same amount of computing and the same amount of money as you did a year ago.”
Despite the focus on inference, one thing missing from today’s announcement is a new image generator. Altman reportedly said in the Code Red memo that image generation will be a key priority going forward, especially after Google’s Nano Banana (the nickname for Google’s Gemini 2.5 Flash Image model) made headlines after its release in August.
Last month, Google launched Nano Banana Pro (also known as Gemini 3 Pro Image). This is an upgraded version with even better text rendering, world knowledge, and an eerie, realistic, unedited feel to the photos. It’s also becoming more integrated across Google’s products, as demonstrated last week by popping 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, but the company did not confirm these plans on Thursday.
OpenAI also announced Thursday that it would be rolling out new safety measures around mental health use and age verification for teens, but did not spend much of its launch time promoting those changes.
This article has been updated with more information about OpenAI’s compute efficiency status.
Do you have confidential information or documents? We report on the inside world of the AI industry, from the companies shaping its future to the people affected by their decisions. Contact Rebecca Bellan (rebecca.bellan@techcrunch.com) or Russell Brandom (russell.brandom@techcrunch.com). For secure communication, you can contact us via Signal at @rebeccabellan.491 and russellbrandom.49.
Source link
