Overview
- Access begins for Plus, Pro, Go, Business, and Enterprise subscribers, with developer endpoints live, while free users are set to get the update later in a phased rollout.
- GPT-5.2 ships in Instant, Thinking, and Pro modes tuned for speed, structured reasoning, and highest‑quality outputs across complex tasks.
- OpenAI reports 70.9% wins or ties versus human professionals on the GDPval knowledge‑work benchmark and 55.6% on SWE‑Bench Pro, with 92.4% on GPQA Diamond.
- The company says error responses drop by roughly 30% versus GPT‑5.1, alongside gains in long‑document handling, tool use for multi‑step workflows, and vision tasks like charts and UI screenshots.
- The release follows an internal ‘code red’ as coverage frames it as a response to Google’s Gemini 3, with outlets noting that many performance figures are company‑reported and will face third‑party scrutiny.