Anthropic’s Claude.ai, API, Claude Code, Console, and Claude for Government all went down for 78 minutes on April 28, 2026 (17:34–18:52 UTC).
Key Takeaways
Full outage window: 17:34–18:52 UTC on April 28; authentication errors hit both claude.ai login and API token flows simultaneously.
Affected surface area was unusually broad: claude.ai, platform.claude.com, api.anthropic.com, Claude Code, Claude Cowork, and Claude for Government in one incident.
Timeline from investigation to resolution: ~78 minutes, with public status updates at 17:41, 17:51, 18:33, and 18:59 UTC.
Root cause was not disclosed; the final update noted success rates returned to normal and teams were monitoring to prevent recurrence.
Hacker News Comment Review
Reliability has become a real enterprise concern: at least one commenter cited $200k/month spend at their org and described the recent outage pattern as damaging enough to escalate to executives, pointing to a gap between Anthropic’s pricing tier and its support quality.
The 90-day uptime figure on status.claude.com drew attention, with builders independently calculating the service has dropped to roughly one nine of reliability, a threshold that makes SLA-sensitive production use difficult to justify.
The incident accelerated discussion of multi-model redundancy: several teams described active hedges across Anthropic, OpenAI Codex, and Gemini, and at least one org reported self-hosting open models on H100s as a direct response to provider downtime risk.
Notable Comments
@SimianSci: $200k/month enterprise spend; executive-level frustration with outage frequency and support quality combined.
@nzoschke: Multi-model, multi-modal tooling (Anthropic, Codex, Gemini; desktop, CLI, web) framed as the new multi-cloud necessity for LLM-dependent teams.
@justrunitlocal: 10-dev org running 8 H100s on open models; cites no downtime and high tok/s as justification over managed API cost.
