Cut reasoning-token weight. Use when thinking or reasoning tokens dominate the bill, when asked to cap a thinking budget, lower effort, or stop simple tasks from burning extended thinking. Light tasks get light thinking: matches thinking budget to task weight, then verifies accuracy held before keep
Cut reasoning-token weight. Use when thinking or reasoning tokens dominate the bill, when asked to cap a thinking budget, lower effort, or stop simple tasks from burning extended thinking. Light tasks get light thinking: matches thinking budget to task weight, then verifies accuracy held before keeping any cap. Never caps blind.