Date: December 11, 2025
OpenAI has officially released GPT-5.2, a groundbreaking new model series designed specifically to dominate professional knowledge work and complex, long-running agentic tasks. Rolling out today for ChatGPT Plus, Pro, and Enterprise users, as well as developers via the API, GPT-5.2 represents a massive leap forward in general intelligence, coding, and reasoning.
The headline feature of this release is the new GPT-5.2 Thinking model, which OpenAI claims is the first AI to perform at or above a human expert level on well-specified professional tasks.
Breaking the “Expert” Barrier: GDPval Benchmarks
The most significant metric from this release is the model’s performance on GDPval, a benchmark evaluating tasks across 44 distinct occupations (including sales, accounting, and manufacturing).
GPT-5.2 Thinking achieved a 70.9% win rate against industry professionals, a staggering increase from the 38.8% score held by the original GPT-5. According to expert judges, the model produced deliverables—such as complex spreadsheets and strategic presentations—at >11x the speed and less than 1% of the cost of human experts.
Three New Tiers: Instant, Thinking, and Pro
The GPT-5.2 release is categorized into three distinct versions to suit different workflows:
- GPT-5.2 Instant: A fast, low-latency “workhorse” model optimized for quick info-seeking, technical writing, and translation.
- GPT-5.2 Thinking: The flagship model for deep reasoning. It excels at multi-step projects, reducing hallucinations by 30% compared to GPT-5.1.
- GPT-5.2 Pro: The powerhouse for scientific and mathematical research, achieving 93.2% on the graduate-level GPQA Diamond benchmark.
State-of-the-Art Coding and Agentic Capabilities
For developers, GPT-5.2 creates a new paradigm in “agentic” coding—where the AI acts as an autonomous engineer rather than just a chatbot.
- SWE-Bench Pro: GPT-5.2 Thinking scored 55.6% on this rigorous software engineering benchmark, which tests capabilities across four languages and real-world repositories.
- SWE-bench Verified: The model reached a new high of 80.0%.
- Front-End Engineering: Early testers report that GPT-5.2 is significantly better at complex UI work, capable of generating interactive 3D elements (like ocean simulations) from a single prompt.
Unmatched Vision and Long-Context Mastery
Professional workflows often involve analyzing massive documents or interpreting visual data. GPT-5.2 sets a new standard here as well:
- Long Context: The model achieved near 100% accuracy on the OpenAI MRCRv2 “needle-in-a-haystack” evaluation, even when processing up to 256k tokens.
- Vision: Error rates on chart reasoning and software interface understanding have been cut in half. The model can now accurately identify components on a motherboard or interpret dashboard screenshots with high spatial awareness.
Availability
GPT-5.2 Instant, Thinking, and Pro are rolling out starting today, December 11, 2025.
- ChatGPT: Available immediately for users on paid plans (Plus, Team, Enterprise).
- API: Developers can access the models now for building next-generation applications.
With improvements that span from abstract reasoning (ARC-AGI-2) to practical customer support (Tau2-bench), GPT-5.2 is positioning itself not just as a tool, but as a reliable, expert-level partner for the modern workforce.
