OpenAI Launches GPT-5.2 with Three-Tiered Architecture for Professional Workflows

OpenAI officially released GPT-5.2 on December 11, 2025, introducing a triple-variant model system designed specifically for professional knowledge work and autonomous agent applications. The company positions this release as its most capable model series to date, with measurable advances in spreadsheet generation, presentation building, coding, vision processing, and extended-context reasoning.

​Three Specialized Variants

The GPT-5.2 lineup consists of Instant, Thinking, and Pro configurations, each optimized for different complexity levels and use cases. Instant prioritizes speed for routine queries like information-seeking, writing, and translation, while Thinking excels at complex structured work such as coding, analyzing long documents, math, and planning. Pro represents the highest-capability tier, designed to deliver maximum accuracy and reliability for difficult problems.

​Benchmark Performance

GPT-5.2 Thinking achieved 100% accuracy on the AIME 2025 competition mathematics examination, establishing a new benchmark record. On GDPval—an evaluation measuring knowledge work spanning 44 occupations—the model wins or ties against industry professionals in 70.9% of direct comparisons, marking the first time an OpenAI model has reached expert-level performance on economically valuable tasks. On OpenAI's benchmark charts, GPT-5.2 Thinking edges out Gemini 3 and Anthropic's Claude Opus 4.5 in nearly every listed reasoning test, from SWE-Bench Pro to GPQA Diamond and ARC-AGI suites.

​Coding and Software Engineering

On SWE-Bench Pro, a real-world software engineering evaluation testing multiple programming languages, GPT-5.2 Thinking scored 55.6%, surpassing GPT-5.1's 50.8%. The model also achieved 80% on SWE-bench Verified. Coding startups like Windsurf and CharlieCode report state-of-the-art agent coding performance and measurable gains on complex multi-step workflows.

​Reliability and Context Handling

GPT-5.2 Thinking responses contain 30% fewer errors than GPT-5.1, making the model more dependable for day-to-day decision-making, research, and writing. The model supports context windows extending to hundreds of thousands of tokens, with near-perfect accuracy on multi-round co-reference resolution tasks. Vision capabilities also improved substantially, cutting error rates approximately in half on chart interpretation and software interface understanding benchmarks.

​Strategic Context

The release comes amid intensifying competition with Google's Gemini 3, which currently tops LMArena's leaderboard across most benchmarks except coding. Earlier this month, CEO Sam Altman reportedly released an internal "code red" memo to staff amid ChatGPT traffic decline and concerns about losing consumer market share to Google. The memo called for shifting priorities, including stalling on commitments like introducing ads and instead focusing on creating a better ChatGPT experience.

​Availability

GPT-5.2 is rolling out to paid ChatGPT subscribers (Plus, Pro, Business, Enterprise) and is immediately available through the API in all three variants. OpenAI also announced new safety measures around mental health use and age verification for teens, though these were not emphasized during the launch briefing.