OpenAI on Thursday, December 11, 2025, unveiled GPT-5.2, calling it the company’s most advanced AI model yet—and its strongest fit for everyday professional work.
OpenAI says GPT-5.2 improves on earlier versions across practical tasks like building spreadsheets and presentations, writing and debugging code, understanding longer context, and handling image-based inputs. The model is available in the API starting Thursday, with ChatGPT access rolling out beginning with paid plans.
The release lands just weeks after OpenAI shipped GPT-5.1, and amid a fresh wave of competition from rivals. Google introduced its Gemini 3 model in November, and Anthropic launched Claude Opus 4.5 around the same time—moves that, according to reporting, helped trigger a “code red” push inside OpenAI to prioritize ChatGPT improvements and pause lower-priority efforts.
“Code red” and the race for mindshare
Speaking to reporters, Fidji Simo, OpenAI’s CEO of Applications, said the “code red” initiative was meant to focus resources and clarify what gets deprioritized, not to suggest GPT-5.2 was rushed out the door.
OpenAI CEO Sam Altman also told CNBC that Google’s Gemini 3 release ultimately affected OpenAI’s metrics less than expected, adding that he anticipates the company will be able to exit the “code red” posture by January.
Instant, Thinking, and Pro: how OpenAI is packaging GPT-5.2
OpenAI is offering GPT-5.2 in three versions:
- Instant: optimized for speed, writing, and information-seeking
- Thinking: tuned for structured work such as planning and coding
- Pro: positioned as the most accurate option for difficult questions
The company is framing the lineup as a “right tool for the job” approach: fast responses when you need them, deeper reasoning when work gets complicated.
Benchmarks and the “which test matters” debate
OpenAI says GPT-5.2 delivers top-tier results across several major AI benchmarks, including:
- SWE-Bench Pro, an agentic coding benchmark
- GPQA Diamond, a graduate-level scientific reasoning benchmark
- GDPval, an OpenAI-created evaluation where the company says GPT-5.2 beat or tied top professionals on 70.9% of well-specified tasks
Notably, the competitive picture is mixed depending on which yardstick you care about. Anthropic says Opus 4.5 scores higher than GPT-5.2 on SWE-Bench Verified (a different SWE-Bench variant), and OpenAI has pushed back by arguing that SWE-Bench Verified is less “contamination resistant” and less representative of real industrial work than SWE-Bench Pro.
Big stakes, bigger expectations
The GPT-5.2 rollout underscores how much OpenAI is betting on its flagship product line. The company is trying to stay ahead in a market where AI tools are rapidly becoming default fixtures in business workflows—and it’s doing so while facing intense scrutiny over the costs of scaling.
OpenAI has been reported to carry a roughly $500 billion valuation, alongside around $1.4 trillion in long-term infrastructure and compute commitments tied to building out the capacity required for next-generation AI systems.
And the audience is already massive: Altman has said more than 800 million people use ChatGPT each week—an adoption curve almost unheard of for a product that only went mainstream after ChatGPT’s 2022 debut.