I'm curious to know if Anthropic mentions anywhere that they use speculative decoding. For OpenAI they do seem to use it based on this tweet [1].

[1] https://x.com/stevendcoffey/status/1853582548225683814