Claude has gone down 164 times in 2026 — what that means for teams running production workflows on it
Anthropic's status page shows 90-day uptime of roughly 99.3–99.4% across claude.ai, the Claude API and Claude Code — well short of the 99.9% enterprise software contracts typically require — after a 7.5-hour outage on 5 August marked the 164th disruption of the year.
6 August 2026
Anthropic’s own status page puts 90-day uptime at roughly 99.36% for claude.ai, 99.42% for the Claude API, and 99.34% for Claude Code. Those numbers sound close to “always available” until you convert them: at that rate, each service is down somewhere between 14 and 23 hours per quarter. A 7.5-hour outage on 5 August 2026, which knocked out multiple Claude model versions, was reported as the 164th disruption Anthropic has logged this year. Anthropic has attributed the pattern to demand growing faster than infrastructure at peak hours — a capacity problem, not a one-off bug.
The number that matters for anyone building on top of this isn’t the outage count on its own, it’s the comparison point: enterprise software contracts typically specify 99.9% uptime or better as a baseline, and Claude’s documented figures sit meaningfully below that across all three services people actually build on.
Why this matters more for coding agents than for chat
An outage on claude.ai is an inconvenience — you wait, or you use something else for an hour. An outage on Claude Code, mid-task, in an agentic workflow that’s mid-refactor or mid-deploy, is a different category of problem: stalled pipelines, half-applied changes, and review queues that back up until service returns. Teams that have built Claude Code into daily engineering workflows this year are, in effect, running production processes against a service with quarter-hour-scale downtime built in, whether they’ve priced that in or not.
So what
This isn’t a reason to avoid Claude Code — it remains one of the strongest tools in the category, and every major AI lab has had a rough infrastructure year keeping pace with demand it didn’t fully plan for. It is a reason to build for the outage you’ll eventually hit rather than assume you won’t: don’t let a single AI coding agent sit as an unacknowledged single point of failure in a critical deployment pipeline, keep a fallback path (a second tool, a manual gate, a pause point) for work that can’t tolerate a multi-hour stall, and ask any development partner leaning heavily on one AI vendor how they’d keep shipping if that vendor went down for an afternoon. That kind of resilience planning is part of how we scope AI-assisted development work — get in touch if you want to talk through what that looks like for your build.