This week, numerous users on X reported unusual behavior in ChatGPT and Codex, including sharper outputs and significantly longer response times, leading to widespread speculation that OpenAI is quietly A/B testing GPT-5.6. Developers shared side-by-side comparisons, with one noting a one-prompt 3D game generation taking over an hour compared to GPT-5.5 Pro's typical 10 minutes. Leaks suggest a knowledge cutoff of December 2025, increased reasoning effort, and improved SVG/3D design capabilities, though OpenAI has not confirmed any testing. The rumors coincide with competitive pressures: China's GLM-5.2 beats GPT-5.5 on the FrontierSWE benchmark, and Anthropic's top models remain pulled due to export controls. OpenAI is reportedly considering price cuts ahead of potential IPOs. Polymarket contracts show an 89% probability of a GPT-5.6 launch between June 22 and June 28.

Key facts

• Users report improved outputs and longer response times in ChatGPT and Codex.
• Developers claim one-prompt 3D games took over 60 minutes vs 10 minutes for GPT-5.5 Pro.
• Leaks suggest a December 2025 knowledge cutoff and increased reasoning effort.
• OpenAI has not confirmed GPT-5.6 testing; Polymarket shows 89% launch probability.
• Rumors emerge amid competition from China's GLM-5.2 and Anthropic's export issues.