Facing mounting rivalry from Google and Anthropic, OpenAI has unveiled a novel AI system, GPT-5.2, which the company asserts outperforms all current models significantly on numerous applications.
The new model, which is being released less than a month after OpenAI debuted its predecessor, GPT-5.1, performed particularly well on a benchmark of complicated professional tasks across a range of “knowledge work”—from law to accounting to finance—as well as on evaluations involving coding and mathematical reasoning, according to data OpenAI released.
TL;DR
- OpenAI released GPT-5.2, a new AI system claiming significant performance improvements over current models.
- GPT-5.2 excels in complex professional tasks, coding, and mathematical reasoning, outperforming competitors.
- The release follows Google's Gemini 3 Pro and OpenAI's internal "code red" to enhance ChatGPT.
- GPT-5.2 also shows advancements in safety features, providing beneficial responses without harmful content.
Fidji Simo, formerly the chief executive of InstaCart and now holding the position of CEO of applications at OpenAI, informed journalists that the model shouldn't be viewed as a direct counter to Google's Gemini 3 Pro AI model, which became available last month. That unveiling led OpenAI CEO Sam Altman to issue a “code red,” postpone the launch of various projects to allocate more personnel and computational power toward enhancing its primary offering, ChatGPT.
“I would say that [the Code Red] helps with the release of this model, but that’s not the reason it is coming out this week in particular, it has been in the works for a while,” she said.
She stated the firm had been developing GPT-5.2 “for many months.” “We don’t turn around these models in just a week. It’s the result of a lot of work,”, she mentioned. This particular model was identified internally with the codename “Garlic,”, as reported by a story in The Information. The preceding day, prior to the model's debut, Altman hinted at its upcoming launch by sharing a video on social platforms showing him preparing a meal heavily featuring garlic.
OpenAI leaders stated that the AI system had been utilized by “Alpha customers”, who assisted in evaluating its capabilities for “several weeks”. This duration would indicate that the AI was finalized before Altman's “code red” announcement.
Among the participants were legal AI venture Harvey, the note-taking application Notion, and the file-management software firm Box, alongside Shopify and Zoom.
OpenAI stated that these users observed GPT-5.2 exhibited a “state of the art” proficiency in employing additional software applications for task completion, alongside superior performance in code composition and error correction.
The deployment of AI models for coding tasks has emerged as a highly competitive area for businesses. While OpenAI initially held a significant advantage, Anthropic's Claude model has gained considerable traction with corporations, reportedly surpassing OpenAI's market presence in some assessments. OpenAI is undoubtedly aiming to persuade clients to re-engage with its AI solutions for coding purposes through the introduction of GPT-5.2.
Simo stated that the “Code Red” was assisting OpenAI in concentrating on enhancing ChatGPT. “Code Red is really a signal to the company that we want to marshal resources in one particular area, and that’s a way to really define priorities and define things that can be deprioritized,” she commented. “So we have had an increase in resources focused on ChatGPT in general.”
The firm also stated that its latest iteration outperforms previous versions in delivering “safe completions”—which they characterize as furnishing users with beneficial responses without uttering statements that could instigate or exacerbate mental health emergencies.
“On the safety side, as you saw through the benchmarks, we are improving on pretty much every dimension of safety, whether that’s self harm, whether that’s different types of mental health, whether that’s emotional reliance,” Simo said. “We’re very proud of the work that we’re doing here. It is a top priority for us, and we only release models when we’re confident that the safety protocols have been followed, and we feel proud of our work.”
The introduction of the latest model occurred concurrently with the filing of a lawsuit a new lawsuit against the firm, which contended that ChatGPT's exchanges with a user experiencing psychological distress played a role in a murder-suicide incident in Connecticut. This firm is also confronting multiple other legal actions asserting that ChatGPT was a factor in individuals' suicides. The company characterized the Connecticut murder-suicide as a “incredibly heartbreaking” and stated its ongoing efforts to enhance “ChatGPT’s training to recognize and respond to signs of mental or emotional distress, de-escalate conversations and guide people toward real-world support.”
GPT-5.2 demonstrated a significant improvement in results on multiple evaluation metrics relevant to business clients. It achieved or surpassed the capabilities of human professionals on numerous challenging work-related assignments, according to OpenAI’s GDPval benchmark, registering at 70.9% of instances. This contrasts with only 38.8% for GPT-5, a system introduced by OpenAI in August; 59.6% for Anthropic’s Claude Opus 4.5; and 53.3% for Google’s Gemini 3 Pro.
Regarding the software development assessment, SWE-Bench Pro, GPT-5.2 achieved a score of 55.6%, surpassing its prior version, GPT-5.1, by nearly 5 percentage points and outperforming Gemini 3 Pro by over 12%.
Aidan Clark, OpenAI's vice president of research (training), opted not to address specific inquiries regarding the precise training techniques employed to enhance GPT-5.2's capabilities. Nevertheless, he indicated that the organization had achieved widespread advancements, encompassing the foundational pretraining phase crucial for AI model development.
Last month, upon unveiling its Gemini 3 Pro model, Google's researchers also announced advancements in both pretraining and post-training methodologies. This development caught some experts by surprise, as they had assumed AI firms had nearly depleted the potential for significant gains from the initial model-building phase. It was further suggested that OpenAI might have been taken aback by Google's achievements in this specific domain.
