The release date is not confirmed. But the pattern is hard to ignore.
Android Authority and multiple developer forums have pointed to June 23 as the expected launch window . Prediction market Polymarket currently prices a June 22-28 release at 78% . OpenAI Chief Scientist Jakub Pachocki has told employees that GPT-5.6 will be a “significant improvement” over GPT-5.5 .
The leaked specs: a 1.5 million token context window and an API price rumored to be one-third of Claude Fable 5 . The model has been spotted in Codex backend logs since mid-May under codenames like iris-alpha, ember-alpha, and kindle-alpha, with kindle-alpha reportedly selected as the release candidate .
Anthropic released Fable 5 on June 9 with a two-week free trial. That trial ends on June 23. OpenAI appears to be waiting for Fable’s trial users to face a decision — subscribe or leave — then offering them a cheaper alternative with a larger context window . If the timing holds, this isn‘t a product launch. It’s a market intervention.
The Contradiction
Here‘s where the story gets interesting. The early test data doesn’t agree on what GPT-5.6 actually is.
One narrative says it beats Mythos. Developer Mark Kretschmann posted on X that GPT-5.6 “very powerful” and beats Mythos on multiple agentic coding benchmarks . The buzz around the model has centered on two upgrades: front-end and UI generation (reportedly strong enough that you don’t need complex prompts to get clean, production-grade interfaces) and visual understanding and image reference tasks .
The other narrative says it‘s a step backward. Developer “Leo” tested both kepler and kindle with the same prompt at the same setting. The result: kindle actually performed worse than kepler . His conclusion: “kindle has regressed compared to kepler. In its current form, it will be easily defeated by Mythos” . The situation has become murky enough that kindle has since been removed from Arena, and a new model called Levi has appeared — though some investigators suspect Levi may be from Meta, not OpenAI .
Both of these claims cannot be fully true. But they can both be partially true. GPT-5.6 may be stronger on some tasks (front-end, UI) and weaker on others (general reasoning, coding complexity). The question is which capability profile matters more to actual users — and whether OpenAI ships the version that wins the benchmark or the version that wins the wallet.
The Price Factor
Even if GPT-5.6 falls short of Mythos on raw performance, pricing could still make it competitive.
Fable 5 is priced at $10 per million input tokens and $50 per million output tokens — roughly twice the price of Opus 4.8 . GPT-5.5 currently charges $5 and $30. Leaks suggest GPT-5.6 could be priced at roughly one-third of Fable 5 . The Wall Street Journal has also reported that OpenAI is considering significant token price cuts to compete with Anthropic .
The reasoning is straightforward. If GPT-5.6 performs close to Mythos on the tasks that matter most and costs substantially less, price becomes the deciding factor for enterprise buyers — especially at a moment when Fable 5’s availability is tangled in regulatory uncertainty and some developers are already looking for alternatives.
What This Actually Means
OpenAI has not confirmed any of this. The leaks are credible — multiple sources, internal logs, and developer tests — but not official. The 78% Polymarket probability is not 100%. The pricing leak could change. The performance claims haven‘t been independently verified across a full benchmark suite.
But the pattern is consistent. GPT-5.6 is coming, it’s aimed at Fable 5, and it‘s timed to land when Fable’s users are most vulnerable.
P.S. The contradiction in the early test results is the most telling detail. One developer says kindle crushes Mythos. Another says it‘s worse than the checkpoint it replaced. The truth probably lies somewhere in the middle — but the market won’t wait for consensus. June 23 is less than a week away. If GPT-5.6 ships on that date, the debate won‘t be settled by benchmarks. It’ll be settled by whether developers actually switch.
