GPT-5.6 - A new standard for intelligence and efficiency
by•
GPT-5.6 is rolling out in the API!
Our new family of models gives builders three options: Sol (our flagship model for the hardest agentic tasks), Terra (balanced for everyday workflows), and Luna (fast and cost-efficient). GPT-5.6 also introduces Programmatic Tool Calling, Multi-agent (beta) for parallel execution, and explicit prompt caching.
On July 23, we're teaming up with Product Hunt for OpenAI Day. The top five launches will each receive $10K in API credits.
Replies
Does it automatically perform semantic compression during preprocessing to reduce input token usage?
@naveenm7 Honesty doesn't let say it definitively, however, semantic compression at the preprocessing stage is not something that they have specified in what has been provided here (Programmatic Tool Calling, Multi-agent, explicit prompt caching being the only named components). It may well be either an undocumented feature or not included in this version of the product. Better to clarify it in the thread, as explicit prompt caching means that they are going more towards developer-driven optimization than automatically altering input which, in turn, does not make sense from the point of view of silent compression.
The API documentation is genuinely some of the best I've worked with, clear examples that actually run and sensible defaults that save hours of trial and error.
Huge improvement. I keep jumping between Claude Code and Codex for development tasks, but I've been stuck using Claude for a few months now. Just a quick test using 5.6 today is making think I should probably switch back to Codex as my main coding AI. It did a very fast and very comprehensive code review this morning and caught a lot of pretty major things that Claude Opus missed when it told me my app was "launch ready".
Sol 5.6 is an extremely powerful update. I used it for a plan for a new software feature that we are building and it nailed everything in the first go. Same prompt with Fable 5 and there were 3 things that I had to push it to change. Don't get me wrong - both of these models are fantastic but it seems that Sol 5.6 just 'gets it'.
Initial impressions are that this is a very capable model and now is my default one on codex.
How do you deal with possible conflicts that might arise due to concurrent writes of the same shared state when Multi-Agent (beta) executes the task in parallel? Is there any locking or merge functionality, or is it up to the developer to manage that?