Overview
- GPT-5.1 debuts in two variants: Instant focuses on warmer, more instruction-following conversation, while Thinking targets complex tasks with clearer answers and variable pacing.
- Both models introduce adaptive reasoning that adjusts thinking time to query difficulty, with OpenAI reporting gains on AIME 2025 and Codeforces benchmarks.
- OpenAI is phasing access to Pro, Plus, Go, and Business users first, Enterprise and Education receive a seven-day early-access toggle, and API availability is slated for later this week.
- ChatGPT adds expanded personality presets including Professional, Candid, and Quirky, plus new controls to tune conciseness, warmth, scannability, and emoji frequency, with a style fine-tuning experiment in settings.
- OpenAI’s system card notes slight regressions on several internal safety tests for GPT-5.1 versus GPT-5, and charts indicate the Thinking model may use substantially more tokens on the toughest queries.