1 nota · última: 20 Jan 2026
The page says `ultrathink` is deprecated, extended thinking is now enabled by default, and 64K-output models can use `MAX_THINKING_TOKENS=63999` for twice the default budget.