The Revise agent is 30% more efficient
We've made the Revise Agent's orchestration more efficient. Based on internal repeated test runs, it now completes the same work using about 30% less tokens, with no drop in result quality. You don't need to change how you work: requests will simply complete faster and use less of your monthly AI limit, so you can get more done with the same plan.
On a similar note, the default model on Revise – GPT 5.6 Luna – had an 80% price cut last week. We started passing those savings on to our users immediately. GPT 5.6 Luna consistently aces our internal test suite, passing 95%+ of test cases. It's an excellent model and we recommend it for the bulk of editing work.
After this price cut and today's efficiency improvements, Revise users editing with Luna should now be getting about 6-7 times more work done with the same usage limits!