[throwdbaaway]

From a quick testing on simple tasks, adaptive thinking with sonnet 4.6 uses about 50% more reasoning tokens than opus 4.6.

Let's see how long it will take for DeepSeek to crack this.