Anthropic has reduced the first-token p90 latency across its product suite, including Claude Code, web, Cowork, and API services. The updates were announced on October 17, 2026, following an internal development sprint that involved over 3,000 performance improvements.
For local Claude Code sessions, the p90 latency decreased from 4.07 seconds to 1.33 seconds. Cloud sessions experienced a reduction from 16.99 seconds to 6 seconds, representing a speed increase of approximately 2.8x.
Improvements were also recorded for the web and Cowork interfaces, where response times dropped from 3.36 seconds to 2.17 seconds. The API first-token response time improved from 1.53 seconds to 1.01 seconds.
The performance gains were achieved through two primary technical changes. First, the company implemented gzip compression for conversation payloads to accelerate data transfer. Second, cloud sessions now begin initializing as soon as a user starts typing, rather than waiting for the prompt submission.
