Overview
- GPT-5.2 debuts with Instant, Thinking, and Pro variants that target speed, deep reasoning, and top‑quality professional outputs, respectively, with priority access for paid tiers.
- OpenAI reports sizable gains over GPT-5.1 across benchmarks such as GPQA Diamond, AIME 2025, FrontierMath, and SWE‑Bench, and says hallucinations drop by roughly 30%–38%.
- In GDPval tests, the Thinking variant is cited by OpenAI as matching or exceeding human professionals in 70.9% of tasks and outperforming Google’s Gemini 3 Pro in that measure.
- Subscription pricing remains unchanged as GPT-5.2 rolls out gradually worldwide, with GPT-5.1 kept as a legacy option for about three months before retirement.
- Early external reactions from AI practitioners are strongly positive, and local outlets report availability in multiple markets including Peru.