Unverified50% confidenceEventExact time
According to Fortune's report on Anthropic's incident postmortem, Claude Code experienced a noticeable capability decline for about a month in March and April: one change caused the model to discard its own reasoning history mid-conversation, and another limited replies between tool calls to 25 words; throughout the entire process the API itself did not change.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/21/2026
First Seen
Valid until: 11/4/2026
Sources
Related Entities
Related Claims
VerifiedThe closer an event is to the cutoff date, the smaller its share in the training corpus, and the less clearly the model has mastered it.66% similarUnverifiedThe author cites a popularized version of the Ebbinghaus forgetting curve, saying that more than half of new information is lost within an hour, and that verbal instructions are basically gone by the next day.64% similarUnverifiedClaude exhibits instruction decay in long conversations, often ignoring instructions written in CLAUDE.md after just a few messages.64% similarUnverifiedThe community suspects that a recent Claude Code update may have adjusted the backend usage calculation logic, causing the same operations to consume more quota.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1080121API
curl https://kongchang.com/api/v1/knowledge/claims/1080121MCP
get_claim(id=1080121)