Anthropic makes claude.ai ~3x faster in two-week sprint; partial outage hits Claude services six days later

Anthropic engineers cut core claude.ai and Claude desktop latency roughly 3x in a two-week August sprint (3,000+ changes, no customer-facing incident or rollback); on Sep 29, 2026 a partial outage raised error rates across claude.ai, the Claude API, Claude…

Anthropic engineers reported making the core experience of claude.ai and the Claude desktop app about three times faster during a two-week August sprint, using Claude itself (via Claude Tag with an internal model roughly comparable to Opus 5.5) to find and ship fixes. At the 75th percentile, time to a typeable page on a fresh load fell from 3.1 seconds to 0.55 seconds, starting a Claude Code session fell from 0.8 to 0.3 seconds, and loading a Claude Cowork cloud session fell from 2.6 to 0.73 seconds. The team focused on four journeys covering about 95% of activity and estimated the changes save tens of thousands of user-hours of waiting per day. More than 3,000 changes were merged with no customer-facing incident or rollback. Separately, on Sep 29, 2026, Anthropic's status page reported a partial outage with elevated error rates across claude.ai (including desktop and mobile apps), platform.claude.com, the Claude API (api.anthropic.com), Claude Code, and Claude Cowork, with users seeing failed requests, conversation load or send errors, and repeated sign-in prompts. A mitigation posted at 14:41 UTC on Sep 29 substantially reduced errors, but some sign-ins, account actions, and Claude Code and Cowork sessions were still failing while a full fix remained in progress.

• Anthropic made claude.ai and the Claude desktop app roughly 3x faster in a two-week August sprint.
• At p75, time to a typeable page on fresh load fell from 3.1s to 0.55s.
• New Claude Code sessions dropped from 0.8s to 0.3s; Cowork cloud session load fell from 2.6s to 0.73s.
• The optimization focused on four journeys covering about 95% of activity, estimated to save tens of thousands of user-hours of waiting per day.
• More than 3,000 changes were merged via Claude Tag, using an internal model roughly comparable to Opus 5.5, with no customer-facing incident or rollback.
• On Sep 29, 2026, a partial outage caused elevated error rates across claude.ai (desktop and mobile apps), platform.claude.com, the Claude API, Claude Code, and Claude Cowork.
• A mitigation at 14:41 UTC on Sep 29, 2026 substantially reduced errors, but sign-in, account actions, and some Claude Code and Cowork sessions were still failing; Anthropic said a full fix was in progress.