1.5 million tokens of super intelligent agents. Altman stirs up Silicon Valley. With a god-level minimalist UI and a 1.5M context window, GPT-5.6 is set to launch urgently next month. The AI battle in June has already erupted ahead of schedule.
Internal codename iris-alpha. Discovered in OpenAI Codex backend logs. The most significant OpenAI leak since GPT-5.
GPT-5.6 demonstrates what developers call a god-level ability. In the default state without any detailed prompts for UI guidance, it independently generates minimalist applications with mature grid layouts, restrained color philosophy, and intuitive navigation — abandoning the high-saturation colors favored by early AIs for a lavender-tinted key component combination. Font weight hierarchy and pixel-perfect spacing make the visual focus extremely clear. This is the end of AI-generated "slop" code.
Probe tests confirm GPT-5.6 context window reaches an astonishing 1.5 million tokens — a 43% increase over GPT-5.5's 1.05M. Extreme tests show smooth responses at 900K input, with perfect handling of requests exceeding 1.05M tokens.
OpenAI will adopt a two-tier approach. The standard version focuses on a leap in multi-step reasoning ability. The Pro version emphasizes enhancement of the intelligent agent workflow. Altman is building a super intelligent agent that can take over all digital living spaces.
Three mysterious internal test code names have been revealed: iris-alpha (Iris), ember-alpha (Ember), and beacon-alpha (Beacon). First batch of checkpoints have started internal testing. The underlying model is already used internally at OpenAI for daily debugging.
OpenAI has compressed its model iteration cycle to just 30-45 days. From GPT-5 (Aug 2025) to GPT-5.6 (June 2026), the cadence has accelerated dramatically — a qualitative change in reinforcement learning cycles and model architecture optimization.
Developer Haider discovered gpt-5.6 mapped in Codex routing logs. What started as a blip became a full-blown confirmation.
Well-known developer Haider, while reviewing Codex's routing logs, noticed a mapped entry for gpt-5.6 among numerous gpt-5.5 call records. Initially dismissed as a canary test or system bug, it quickly disappeared.
Multiple developers successfully called the unannounced GPT-5.6 model in the Codex environment through ChatGPT Pro's OAuth authentication. The model runs on openai/gpt-5.6 with configurable inference levels.
Well-known leaker Leo confirmed: "The development of GPT-5.6 has been fully promoted. First batch of checkpoints have started internal testing. Expected official release next month." Alongside came the three codenames: iris, ember, beacon.
Polymarket odds for GPT-5.6 release before June 30 have soared to over 85%. The betting market reflects the overwhelming confidence in an imminent release that has the entire AI community watching.
From GPT-5 to GPT-5.6 — the release cadence has compressed from years to weeks.
June 2026 will be a feast of large models. Anthropic, Google, and Musk are all preparing their counter-strikes.