GPT-5.5 has only been out for 3 weeks! GPT-5.6 has already been integrated into Codex, with context size increased to 1.5M. Multiple sources have leaked that it will be released in June.

According to DataPulse Beating monitoring, just three weeks after the release of GPT-5.5, the successor GPT-5.6 has already been successfully run by external developers. Several developers, using ChatGPT Pro OAuth authentication, have successfully invoked the yet-to-be-announced gpt-5.6 model in the Codex environment. Probe tests have shown that the context window has reached 1.5M tokens, representing an approximately 43% increase from the GPT-5.5 API's 1.05M.

The first trace of GPT-5.6 appeared on April 28th. Developer Haider, while reviewing Codex route logs, discovered that the majority of calls pointed to gpt-5.5, but there was one mapping explicitly stating gpt-5.6. He later revised his assessment, suggesting it seemed more like a canary test or a bug as the entry quickly disappeared.

However, a significant change has been observed this week. Some developers who encountered an "model is not supported" error when specifying gpt-5.6 last week were able to proceed this week directly using Pro's OAuth. Most importantly, the surge in the context window is notable: while GPT-5.5's API context is 1.05M tokens, going through Codex OAuth reduces it to only 400K. In contrast, GPT-5.6's probe directly reached 1.5M tokens, nearly a 1.5x increase. Developer tests in OpenCode have also confirmed that the model responds normally even above 900K tokens, and requests above 1.05M are equally accepted. The model claims to be running on openai/gpt-5.6 in conversations, the inference level can be set to xhigh, fast mode is available, and the speed is fast.

Blogger Leo announced today that the development of GPT-5.6 has progressed comprehensively, with the first batch of checkpoints starting internal testing over the past few days, expected to be released next month, while exposure of two internal codenames, ember-alpha and beacon-alpha.

Haider analyzed OpenAI's iteration cadence: from annual updates to biannual, quarterly, bimonthly, now shortened to 30-45 days, from which it is speculated that GPT-5.6 should be released in early June. He also predicts that GPT-5.6 will surpass GPT-5.5 on several benchmarks where Mythos still leads. The reason being that GPT-5.5 is already very close, and with more reinforcement learning, the gap can be widened, especially with OpenAI's stronger RL loop in coding, mathematics, and scientific research.

The probability, as per Polymarket, of GPT-5.6 being released by June 30th is currently around 85%.