Source record

Published by Amp News (ampcode.com) (T1 source). The original is at https://ampcode.com/news/gpt-5.5.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Amp's deep coding mode now runs on GPT-5.5, which the team found more steerable and better at the agent loop than GPT-5.4. They lowered the default reasoning effort from high to medium after internal evals, and share prompting patterns emphasizing outcomes, constraints, and verification steps.

The internal eval result that higher reasoning effort cost more and performed worse is a useful, counterintuitive data point; the prompting patterns and Opt+D effort toggle are directly reusable.

Cited by

No citations on record.