Overview
- Availability starts with paid ChatGPT plans and developers, with Instant, Thinking, and Pro variants in a staged rollout while GPT-5.1 remains as a legacy option for three months.
- OpenAI says GPT-5.2 sets new highs on internal and public tests, topping suites like SWE-Bench Pro and GPQA Diamond and beating or tying industry professionals on 70.9% of GDPval tasks.
- The company reports fewer mistakes and safer behavior, including 38% fewer hallucinations for the Thinking model on factual benchmarks and reduced response errors.
- Enterprise alpha testers such as Notion, Box, Shopify, Harvey, Zoom, and Databricks trialed the model for long‑context work, agentic tool use, and stronger coding performance.
- The launch follows competitive pressure from Google and Anthropic; Sam Altman says the 'code red' focus should wind down by January, and limited age‑prediction trials are underway ahead of a planned adults‑only mode in Q1 2026.