GPT 5.6 coming soon...
OpenAI’s GPT-5.6 remains unannounced and unreleased as of today, but the speculation is heating up fast.
GPT-5.5 (launched late April) is still the current public flagship, yet internal traces of GPT-5.6 already appeared in Codex backend logs back in mid-May.
Internal codenames like iris-alpha have been spotted.
Chief scientist Jakub Pachocki reportedly told staff it will be a “meaningful improvement” over GPT-5.5.
Prediction markets are pricing an 80-89% chance of a public release before the end of June.
Some reporting suggests it could drop very soon and may coincide with a major ChatGPT overhaul.
Rumored upgrades include:
A jump to 1.5 million token context (vs ~1M in the current GPT-5.5 API). This would be aimed at full codebases and very long documents.
Stronger multi-step agentic workflows, better error recovery, and improved frontend/UI generation — an area where many feel GPT-5.5 still has headroom.
Gains in efficiency, safety guardrails, and overall reasoning.
It looks like OpenAI is preparing for a price war on the API side to stay competitive.
Meanwhile, Anthropic’s Claude Fable 5 (their new Mythos-class model) launched publicly around June 9 and was getting strong early feedback for software engineering and complex agentic tasks.
It appeared to outperform or match GPT-5.5 and even Opus 4.8 in some coding benchmarks during its short window.
Then, just three days later on June 12, Anthropic was forced to suspend access globally after receiving a U.S. government export control directive citing national security concerns.
The order restricted use by foreign nationals (including staff), so Anthropic pulled Fable 5 and Mythos 5 for everyone to comply. It was a regulatory move, not a technical failure.
Developers are already extracting Fable 5’s ~120k-character system prompt and applying it to Opus 4.8 as a workaround.
So the question many are asking: Will GPT-5.6 outperform the now-removed Claude Fable 5?
It’s still speculative.
GPT-5.6 is explicitly being built to compete in this exact environment.
If the 1.5M context and frontend/agentic improvements materialise, combined with OpenAI’s ecosystem and potential aggressive pricing, it has a realistic shot at taking the lead in practical coding and long-horizon agent work.
However, recent Claude models have been very strong on careful reasoning and calibrated outputs.
Fable 5 was Anthropic’s aggressive push in that direction.
We’re in a period of extremely rapid iteration.
One model gets pulled for regulatory reasons while the next challenger is already in internal testing.
This space is moving faster than most people realise. Worth keeping a close eye on the next 2–3 weeks.