Claude's extended thinking

Summary

Claude 3.7 Sonnet introduces extended thinking, letting users toggle deeper reasoning and developers set a thinking budget. The same model spends additional computation on sequential reasoning, while agent training supports iterative function calls, computer use, and open-ended tasks. The post examines visible thought processes, noting benefits for trust and research but also limitations around faithfulness, safety, and jailbreak risk; it therefore treats the feature as a research preview. Reported examples include stronger OSWorld performance, playing Pokémon Red through tens of thousands of interactions and progressing through three Gym Leaders, plus logarithmic math gains as thinking tokens increase. Safety work includes encrypted thoughts for some harmful content and computer-use prompt-injection mitigations that raise prevention from 74% to 88%.

Takeaways

1. 1

   Claude 3.7 Sonnet’s computer-use prompt-injection mitigations prevent attacks 88% of the time, compared with 74% without those mitigations.

2. 2

   Visible thought processes may help users inspect answers and support alignment research, but the post says current thinking traces are not reliable evidence of model safety because decisions may depend on factors not explicitly discussed.

3. 3

   With memory, screen-pixel input, and button-press function calls, Claude 3.7 Sonnet played Pokémon Red continuously through tens of thousands of interactions and defeated three Gym Leaders.