• OpenAI releases GPT-5.2 as pressure from Google and Anthropic grows.
• GPT-5.2 improves performance while OpenAI weighs safety and user trust.

Pressure inside OpenAI has been building as competition around AI models tightens. That strain came into view this week with the release of GPT-5.2, a new model that the company says improves writing, coding, and reasoning, and follows an internal “code red” aimed at sharpening ChatGPT’s performance.

The internal push was meant to focus attention, rather than rush a release, according to OpenAI executives.

Competition tightens as OpenAI rolls out GPT-5.2

“We announced this code red to really signal to the company that we want to marshal resources in one particular area, and that’s a way to really define priorities,” said Fidji Simo, OpenAI’s CEO of applications, during a briefing with reporters on Thursday. “We have had an increase in resources focused on ChatGPT in general.”

Simo said the work on GPT-5.2 had been underway for months and was not moved forward because of the code red. Still, she acknowledged that shifting more people and effort toward ChatGPT has been “helpful.” The release comes less than a month after GPT-5.1, reflecting a faster cadence as rivals continue to gain ground.

When ChatGPT first launched in 2022, OpenAI’s position at the top of the market looked secure. That advantage has narrowed. Google has emerged as one of the strongest challengers, with its Gemini 3 model drawing a positive response in the tech industry. Its Gemini app has grown quickly, reaching more than 650 million monthly active users. OpenAI, by comparison, says ChatGPT now has about 800 million weekly active users.

Pressure is also coming from Anthropic, whose Claude models have gained traction with enterprise customers, particularly for coding tasks. Some industry figures suggest Claude has overtaken OpenAI in parts of the enterprise market, raising the stakes for GPT-5.2’s performance in software development.

Competition has forced OpenAI to reassess its priorities. Some longer-term ideas, like putting ads in ChatGPT, have taken a back seat while the company concentrates on improving its core models and user experience.

Benchmark gains and early testing results

GPT-5.2 follows the same tiered approach as recent OpenAI releases. The Instant version is designed for faster replies and basic information tasks, and Thinking targets more complex work, including coding, math, and planning. Pro sits at the top, aimed at users who need higher accuracy on difficult questions.

OpenAI positions GPT-5.2 as its strongest option so far for everyday professional tasks. On GDPval, an internal benchmark that compares AI models with human professionals in 44 occupations, GPT-5.2 Thinking achieved the highest scores the company has recorded. OpenAI says the model matched or exceeded human expert performance in just over 70% of tasks, outpacing earlier OpenAI models as well as recent releases from Google and Anthropic.

The company also points to gains in software development. On SWE-Bench Pro, a benchmark focused on real-world coding tasks, GPT-5.2 scored higher than its immediate predecessor and outperformed Gemini 3 Pro. OpenAI says the model also shows stronger ability to use external software tools to complete tasks, based on feedback from early testers.

Those testers included a group of “alpha customers” that had access to the model for several weeks before launch, like legal AI startup Harvey, note-taking app Notion, file-management firm Box, and well-known companies that included Shopify and Zoom.

OpenAI also highlights progress on accuracy. Max Schwarzer, the company’s post-training lead, said the new model shows a meaningful drop in hallucinations. On benchmarks focused on factual questions, OpenAI says GPT-5.2 Thinking produced 38% fewer hallucinations than GPT-5.1.

The new models are being rolled out to both ChatGPT users and developers through OpenAI’s API. The company says the updates improve performance in routine and more advanced uses.

Growth pressure and ongoing safety concerns

OpenAI’s recent history shows that strong benchmark results do not always match how users feel in practice. When GPT-5 launched earlier this year, many users complained that its responses felt distant and impersonal. The issue was not reflected in standard tests, and OpenAI released an update days later to make the model sound “warmer.”

That episode highlights an ongoing balancing act for the company. OpenAI wants ChatGPT to feel engaging and natural, without pushing it toward excessive agreement or emotional dependence. Over the past year, the company has faced growing scrutiny over mental health and how people rely on conversational AI.

In October, OpenAI released a report showing that more than a million people talk to ChatGPT about suicide each week. Around the same time, a research leader involved in the company’s mental health efforts announced plans to leave, according to internal messages. The company also faces several lawsuits linked to claims about ChatGPT’s role in cases involving severe psychological distress.

Despite such challenges, competition from Google, Anthropic, and Meta has raised the stakes around user growth. In a memo sent in October, OpenAI’s head of ChatGPT, Nick Turley, warned employees that the company was facing “the greatest competitive pressure we’ve ever seen,” according to The New York Times. Turley reportedly set a target to raise daily active users by 5% before 2026.

Alongside the release of GPT-5.2, OpenAI says it has continued to improve how ChatGPT handles sensitive prompts linked to self-harm, emotional distress, or reliance on the model. The company is also beginning to roll out its age-prediction system in some countries. The system is meant to estimate whether a user is under 18 and apply stricter content controls when needed.

Simo said OpenAI plans to introduce an “adult mode” in the first quarter of 2026. Sam Altman has previously said the feature would allow users over 18 to engage in “erotic” conversations with ChatGPT.

Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.

Tech Wire Asia is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.