Overview
- OpenAI has been running internal tests for GPT-5.6 with a model identifier appearing in backend logs in mid-May and chief scientist Jakub Pachocki calling it a “meaningful improvement” in early June.
- Reports and user posts this week describe slower, longer-running ChatGPT and Codex responses that some testers say reflect a Pro build undergoing stealth A/B testing inside the service.
- Rumors point to a dramatically expanded context window of up to 1.5 million tokens for GPT-5.6 but that specification has not been confirmed by OpenAI or supported by public benchmarks.
- Prediction markets have priced a high chance of a release between June 22 and June 28, with Polymarket contracts trading heavily and probabilities in the mid-80s percent, though OpenAI has made no formal announcement.
- If accurate, the Pro variant’s focus on longer runtimes and multi-step agent reasoning would strengthen OpenAI’s position against rivals like Google’s Gemini and could change how developers build long-document, codebase, and planning applications while likely affecting enterprise pricing and API terms.