The advisor tool now supports max_tokens to cap advisor output per call.

Builders can limit advisor responses when workloads do not need full-length output, reducing latency and output token cost.

Set tools[].max_tokens on the advisor definition for workloads that need capped advisor output.

• Jul 13, 2026ChatGPT now lets users delete memory-summary memories and turn memory off from the three-dot menu.
• Jul 13, 2026GET /v1/environments/{id}/work is now available on Claude Platform on AWS for self-hosted sandboxes.
• Jul 13, 2026Codex added rate-limit reset banking and referral invitations for eligible Plus and Pro users.

This is the world we build in. It moves this fast. We move with it.