Prediction markets are giving GPT-5.6 an 80-89% chance of releasing in June 2026. Two versions are expected: GPT-5.6 and GPT-5.6 Pro.
If it ships this month, OpenAI will have released three major models in three months — GPT-5.4, then 5.5, now 5.6 — a cadence no frontier lab has sustained before.
The signal from Washington and the markets
Treasury Secretary Scott Bessent told CNBC on May 14 that he anticipates a “step-function jump” in upcoming model releases from OpenAI and Google. Combined with prediction-market odds near 89%, the expectation is less “if” than “which day.”
The three-month cadence
| Model | Released |
|---|---|
| GPT-5.4 | ~March 2026 |
| GPT-5.5 | April 23, 2026 |
| GPT-5.6 | Expected June 2026 |
What GPT-5.6 likely improves
GPT-5.5 launched April 23 and retook the benchmark crown from Claude on 14 of 18 benchmarks. Following OpenAI’s pattern, GPT-5.6 should push long-horizon reasoning and coding further, with the Pro variant reserved for the heaviest agentic and reasoning workloads. The standard-versus-Pro split lets OpenAI serve cost-sensitive volume and frontier-performance buyers from one release.
The race with Claude
The competitive context is sharp. DeepSWE already showed GPT-5.5 leading Claude by 16 points on long-horizon engineering, while Claude Opus 4.8 reclaimed shorter-task coding. GPT-5.6 versus the anticipated Sonnet 4.8 is the next round of a head-to-head that now turns over roughly every six weeks.
Codex adoption shows the stakes: OpenAI’s coding agent has 5M-plus weekly active users, up from 2M in April. For developers, the practical takeaway is to expect a new frontier option this month and to benchmark by task horizon — short fixes and long multi-file builds still favor different models.
