OpenAI's GPT-5.6 Spotted in Codex Logs with 1.5M Context Window

Developers found an unannounced model, GPT-5.6 (codename iris-alpha), in OpenAI Codex backend logs. It supports a 1.5 million token context window and is expected to launch in June 2026. Other versions like ember-alpha and beacon-alpha were also spotted. Compared to GPT-5.5's 1.05 million token limit, GPT-5.6 boosts capacity by about 43%. In extreme tests, it handled over 90k tokens smoothly and even processed requests beyond 1.05 million tokens. The model also generated a clean note app interface called Lumen Notes with minimal prompts, showing improved front-end coding skills.