Anthropic publishes postmortem on April Claude Code outages
One bug, a caching optimisation meant to run once, instead cleared Claude Code's reasoning on every turn for the rest of a session, silently degrading output quality.
- Models & capabilities
- Colour
Anthropic published a postmortem attributing weeks of degraded Claude Code output quality in April 2026 to three separate, unrelated changes rather than a single fault. A 4 March change had lowered the default reasoning effort from “high” to “medium” to cut latency, a trade-off Anthropic later judged mistaken and reversed on 7 April. A 26 March caching optimisation, intended to clear stale reasoning only from sessions idle for over an hour, instead ran on every turn for the rest of a session, causing Claude to progressively lose context — fixed on 10 April, after Opus 4.7’s code-review tooling caught the bug that Opus 4.6 had missed. A verbosity-limiting system prompt added around the same time cut response quality by a measured 3% in evaluations and was reverted on 20 April. Anthropic said all three issues were resolved by 23 April and reset affected subscribers’ usage limits.
In the commentary
What people were saying around this time — external links, from the record's commentary rail.