Anthropic says Claude Code now starts responding much faster after adding request gzip and session prewarming.
Claude Code compresses the conversation sent on each turn, reducing upload time for large contexts. Cloud sessions begin warming as soon as the user starts typing rather than waiting for submission, according to AICoder.
Anthropic’s reported p90 wait before the first token fell from 4.07 seconds to 1.33 seconds for local sessions and from 16.99 seconds to 6.00 seconds in the cloud. Those measurements exclude model inference time, so the gains come from transport and startup work rather than faster generation.
For builders, the playbook is practical: compress long conversational payloads, prewarm execution environments early, and test proxy compatibility. The changes follow the latency-focused approach described in Anthropic’s broader engineering account.