The rumors were true: OpenAI on Thursday announced the release of its new frontier Large Language Model (LLM) family, GPT-5.2,
It’s a significant moment for the AI pioneer, which has faced intense pressure since rival Google’s Gemini 3 LLM took the top spot on major third-party performance leaderboards and several key benchmarks last month, though OpenAI leaders stressed in a press briefing that the timing of this release was discussed and worked on well in advance of Gemini 3’s release.
OpenAI declares GPT-5.2 as its own "The most capable model series ever for professional knowledge work," Aiming to reclaim the performance crown with significant gains in logic, coding and agentic workflow.
"This is our most advanced Frontier model and the most robust model yet on the market for commercial use." said Fidzi Simo, CEO of Applications at OpenAI, during a press conference today. "We’ve designed 5.2 to unlock even more economic value for people. It is better at building spreadsheets, creating presentations, writing code, understanding images, understanding longer context, using tools, and handling complex, multi-step projects."
GPT-5.2 has a massive 400,000-token context window – allowing it to ingest hundreds of documents or large code repositories at once – and a 128,000 maximum output token limit, enabling it to generate comprehensive reports or complete applications at once.
The model also includes a knowledge cutoff of August 31, 2025, ensuring that it is up to date with relatively recent world events and technical documentation. This clearly includes "logic token support," Popular thought-process series used to verify the underlying architecture "O1" series.
‘Code Red’ reality check
The release comes following Informationreport of an emergency "code Red" A directive from CEO Sam Altman to OpenAI staff to improve ChatGPT – a move reportedly designed to mobilize resources following "quality difference" Exposed by Gemini 3. The Verge Similarly, the timing of the release of GPT-5.2 was also reported before the official announcement.
During the briefing, OpenAI officials acknowledged the directive but rejected the statement that the model was rushed only to respond to Google.
"It’s important to note that this has been in the works for many, many months," Simo told reporters. He clarified that while "code Red" While it helped the company focus, it wasn’t the sole driver of the timeline.
"We actually announced this code red to signal to the company that we want to use resources in a particular area… but that’s not why it’s coming up specifically this week."
Max Schwarzer, head of OpenAI’s post-training team, echoed this sentiment in dispelling the idea of a panic launch. "We’ve been planning this release for a long time… we talked about this specific week several months ago."
An OpenAI spokesperson further clarified that "code Red" The call applies to ChatGPT as a product, not just underlying model development or the release of new models.
Under the Hood: Quick, Thinking, and Pro
OpenAI is splitting the GPT-5.2 release into three different tiers within ChatGPT, a strategy likely designed to balance massive computation costs "logic" Models with user demand for speed:
- GPT-5.2 Immediate: Optimized for speed and daily tasks like writing, translating and retrieving information.
-
GPT-5.2 Thinking: designed for "complex, structured tasks" And long-running agents, this model leverages deep reasoning chains to handle coding, math, and multi-step projects.
-
GPT-5.2 Pro: New Heavyweight Champion. OpenAI describes it as "The smartest and most reliable option," Providing the highest accuracy for tough queries where quality is more important than latency.
For developers, models are immediately available in an application programming interface (API). gpt-5.2, gpt-5.2-chat-latest (immediate), and gpt-5.2-pro,
The numbers: Beating the benchmark
The GPT-5.2 release includes leading metrics in most domains – especially those that target "professional knowledge work" The gap where competitors have gained recently.
OpenAI highlights a new benchmark called GDPval, which measures the performance "well-specified knowledge work tasks" In 44 professions.
"GPT-5.2 thinking is now state-of-the-art on that benchmark… and outperforms or matches top industry professionals on 70.9% of well-specified business tasks, such as spreadsheet, presentation and document creation, according to expert human judges." Simo said.
In the critical area of coding, OpenAI is claiming a decisive lead. On SWE-Bench Pro, a rigorous assessment of real-world software engineering, GPT-5.2 Thinking set a new state-of-the-art score of 55.6%, Schwarzer said.
He emphasized that it is the benchmark "More contamination resistant, challenging, diverse and industrially relevant than previous benchmarks such as SWE-Bench Verified."Other key benchmark results include:
- GPQA Diamond (Science): GPT-5.2 Pro scored 93.2%, beating GPT-5.2 Thinking (92.4%) and GPT-5.1 Thinking (88.1%).
-
FrontierMath: On Tier 1-3 problems, GPT-5.2 Thinking solved 40.3%, a significant jump from the 31.0% achieved by its predecessor.
-
ARC-AGI-1: GPT-5.2 Pro is reportedly the first model to surpass the 90% threshold on this general logic benchmark, scoring 90.5%
price of intelligence
Performance comes at a premium. While ChatGPT subscription pricing remains unchanged for now, API costs for the new flagship models are higher than previous generations, reflecting higher compute demands. "Thinking" Method
- GPT-5.2 Thinking: at cost $1.75 per 1 million input tokens and $14 Per 1 million output tokens.
-
GPT-5.2 Pro: costs increase significantly $21 per 1 million input tokens and $168 Per 1 million output tokens.
The price of GPT-5.2 Thinking in the API is 40% higher than the standard GPT-5.1 ($1.25/$10), indicating that OpenAI views the new reasoning capabilities as a solid value-add rather than just an efficiency update.
The high-end GPT-5.2 Pro follows the same pattern, costing 40% more than the previous GPT-5 Pro ($15/$120). While expensive, it still undercuts OpenAI’s most specialized reasoning model, o1-pro, which remains the most expensive offering on the menu at $150 per million input tokens and $600 per million output tokens.
OpenAI argues that despite the high per-token cost, the model’s "greater token efficiency" And the ability to solve tasks in fewer turns makes it economically viable for high-value enterprise workflows.
Image Building: Nothing new yet…but ‘more to come’
During the briefing, VentureBeat asked OpenAI participants whether the new release included any enhancements to image creation capabilities, noting the excitement about similar features in recent competing launches like Google’s Gemini 3 Image, aka Nano Banana Pro.
Unfortunately for those who want to recreate the text-and-information heavy graphics and image editing capabilities, OpenAI officials clarified that GPT-5.2 comes with no current image improvements compared to the previous GPT-5.1 and OpenAI’s integrated DALL-E 3 and gpt-4o native image generation models.
"At Image General, there’s nothing to announce today, but there’s more to come," Simo said. He acknowledged the popularity of the feature, saying, "We know this is a very important use case that people love, which we’ve introduced [to] market, and so there will certainly be more to come."
Aidan Clark, OpenAI’s head of training, also declined to comment on the specifics of the scene creation, saying, "I can’t really speak to Image General itself."
‘Mega-Agent’ era
Beyond raw scores, OpenAI is positioning GPT-5.2 as the engine for the new generation "long running agent" Able to execute multi-step workflows without human assistance."
Box found that 5.2 could extract information from long, complex documents about 40% faster, and also saw a 40% increase in reasoning accuracy for life sciences and health care." Simo said.
He also noted that Notion reported the model "Performs better than 5.1 in every dimension… and it really excels at the vague, long-simmering tasks that define real cognitive work."Schwarzer said coding startups like Augment Code have found the model "Provided significantly stronger deep code capabilities than any prior model," That’s why it was chosen to power their new code review agent. Visual capabilities have also seen an upgrade.
OpenAI’s release blog post shows an example "A traveler reports a delayed flight, missed connections, an overnight stop in New York and the need for medical sitting."
the outcome? "GPT‑5.2 manages the full range of functions – rebooking, special-assistance seating and compensation – providing more complete results than GPT‑5.1."
A new evaluation called ScreenSpot-Pro, which tests a model’s ability to understand GUI screenshots, finds GPT-5.2 Thinking achieves 86.3% accuracy, compared to only 64.2% for GPT-5.1.
Science and credibility
OpenAI leaders also emphasized the usefulness of the models for scientific research, attempting to take the conversation beyond simple chatbots to research assistants.
Aidan Clark, head of the training team, shared an example of a senior immunology researcher testing the model.
"They tested it by asking them to generate the most important unanswered questions about the immune system," Clark said. "That immunology researcher reported that GPT-5.2 posed more pointed questions and stronger explanations of why these questions matter than any previous PRO model.
"Reliability was another key focus. Schwarzer claims new model "Much less hallucination than GPT-5.1," Bearing in mind that on a set of unknown queries, "There were 38% fewer errors in responses."
‘Vibe’ Shift
Interestingly, OpenAI acknowledged that not every user may immediately like the new model.
When asked why older models like GPT-5.1 would remain available, Schwarzer admitted "The models change a little each time.
"Some users may find that they prefer the vibes of the previous model, even though we think the latest model is generally much better," Schwarzer said. He also said that for some enterprise customers who have "The prompt has actually been fixed for a specific model," It is possible "small regression," Requires access to older versions.
Safety, ‘Adult Mode’ and the roadmap for the future
Addressing security concerns, Simo confirmed that the company is preparing to implement "adult mode" In the first quarter of next year, after the implementation of a new era prediction system.
"We are in the process of improving it," Simo said about age prediction technology.
"We want to do this before launching Adult Mode."Looking ahead, industry reports suggest that OpenAI is working on a more fundamental architectural change under the codename "Project Garlic," Aiming for a major release in early 2026.
Although officials did not comment on specific future roadmaps during the briefing, Simo remained optimistic about the economics of its current trajectory.
"If you look at historical trends, the count has increased almost 3 times every year for the last three years." she explained. "Revenue has also increased at the same pace… this is creating a virtuous cycle."
Clark said efficiency was improving rapidly: "The model we’re releasing today gets even better scores [on ARC-AGI] With approximately 400 times less cost and less computation" Compared to models from a year earlier.
GPT-5.2 Instant, Thinking, and Pro is starting to be available today in ChatGPT for paid users (Plus, Pro, Team, and Enterprise). The company says the rollout will be done gradually to maintain stability.
<a href