Anthropic’s Claude Recovers After 90-Minute Multi-Model Outage

Anthropic’s Claude service recovered on June 22, 2026, after a 90-minute incident caused elevated error rates across several major Claude models. The disruption affected Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 4.6, and Haiku 4.5.

The incident began at 00:37 UTC, when Anthropic said it was investigating elevated error rates. The company identified the issue at 01:11 UTC, restored several models in stages, and marked the incident resolved at 02:06 UTC, according to the official Claude status page.

The disruption mattered because Claude is now used across chat, developer workflows, APIs, and agentic tools. Even a short incident can affect teams that rely on Claude for coding, automation, support, research, and internal productivity workflows.

Five Claude Models Were Affected

Anthropic’s incident entry listed five affected models: Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 4.6, and Haiku 4.5. The company did not publicly describe the root cause in detail, but its updates showed a staged recovery rather than a single instant restoration.

Opus 4.8 recovered first at 01:16 UTC. Haiku 4.5 followed at 01:33 UTC, and Opus 4.7 recovered at 01:56 UTC. Anthropic then implemented a fix and moved into monitoring at 02:01 UTC before declaring the incident resolved five minutes later.

Users also reported Claude access issues on outage-tracking sites. Downdetector’s Claude AI page tracks user-submitted reports across areas such as Claude Chat, Claude API, Claude Code, login, website access, lag, and latency.

Claude Outage Timeline

Time UTC	Status update	What it meant
00:37	Investigation opened	Anthropic began investigating elevated error rates across multiple models.
01:11	Issue identified	The company said a fix was being implemented.
01:16	Opus 4.8 recovered	The first affected model returned to normal success rates.
01:33	Haiku 4.5 recovered	Anthropic continued staged recovery across models.
01:56	Opus 4.7 recovered	Another affected model returned to normal operation.
02:01	Fix implemented	Anthropic moved the incident into monitoring.
02:06	Resolved	The multi-model incident was marked fully resolved.

The full window lasted 89 minutes from the first investigation update to final resolution. That is why the incident is being described as a roughly 90-minute outage.

Anthropic’s public updates did not say that every Claude product was completely offline. The official wording points to elevated error rates, which can mean failed requests, degraded reliability, or intermittent access rather than total unavailability for every user.

However, for developers using the Claude API documentation and platform to build production integrations, elevated error rates can still create visible problems. Failed API calls can interrupt customer-facing apps, automated workflows, and internal tools.

Why the Outage Hit Developers Hard

Claude is no longer only a chatbot for individual users. It now supports developer tooling, enterprise automation, and agent-based workflows, which makes reliability more important for teams that depend on it throughout the workday.

Claude Code is especially relevant for engineering teams because it lets developers work with Claude inside coding environments. When model success rates drop, those workflows can stall even if the rest of the development stack remains online.

Anthropic’s developer documentation positions Claude as a platform for building applications through the API, quickstarts, SDKs, and Claude Code. That makes short outages more operationally important than they were when AI assistants were used mainly for casual chat.

Services and Workflows Potentially Affected

Area	Possible impact during elevated errors	Who feels it most
Claude.ai	Failed prompts, blank replies, or delayed responses	Individual users and business teams
Claude API	Failed requests or retries in connected applications	Developers and SaaS teams
Claude Code	Interrupted coding sessions and blocked automation	Software engineers and DevOps teams
Claude Cowork	Delayed autonomous work on files, apps, and tasks	Knowledge workers and enterprise users
Internal AI workflows	Fallbacks, degraded service, or manual workarounds	Organizations with AI-assisted operations

Claude Cowork adds another reliability consideration because it is designed to handle tasks across local files, applications, and user workflows. Anthropic describes Claude Cowork as an agentic tool that can take a goal and return a finished deliverable.

That kind of workflow can create more visible disruption during service instability. If an AI agent is drafting documents, handling files, or supporting internal operations, elevated errors can delay the whole task rather than just one chat response.

The incident also came during a month with repeated Claude reliability events. Anthropic’s status history shows multiple June incidents affecting Opus 4.8 and other models, including another Opus 4.8 incident later on June 22 that was resolved at 14:44 UTC.

June Has Been a Busy Month for Claude Status Updates

The June 22 multi-model incident was not the only reliability event listed on Anthropic’s status history this month. The status page shows several June incidents involving Opus 4.8, Opus 4.7, Haiku 4.5, Sonnet 4.6, and broader model error rates.

Frequent small incidents do not prove a systemic failure, but they do show that AI platforms now face uptime expectations similar to cloud infrastructure. Developers and companies increasingly treat model availability as part of their own service reliability.

The official Claude status dashboard remains the main source for live updates. Users running critical workloads should monitor it directly rather than relying only on social media posts or user reports.

What Claude Users Should Do After an Outage

Most individual users do not need to take any action after a resolved elevated-error incident. They can usually retry prompts or refresh their session once the service returns to normal.

Developers and enterprise teams should take a more structured approach. They should check failed API calls, review retry behavior, confirm queued jobs completed properly, and look for any workflow that may have silently failed during the outage window.

• Review application logs between 00:37 UTC and 02:06 UTC on June 22.
• Check whether failed Claude API requests were retried successfully.
• Confirm that background jobs and automation tasks completed after recovery.
• Add fallback handling for model errors and timeout responses.
• Monitor the Claude status page for follow-up incidents.
• Use provider redundancy for business-critical AI workflows where possible.

Why AI Outages Matter More Now

AI assistants have moved from optional productivity tools into daily business infrastructure. Many teams now use them for coding, research, document work, customer support, analysis, and internal automation.

That shift changes how outages should be measured. A 90-minute disruption may sound short, but it can affect deployments, support queues, sprint work, and time-sensitive business tasks when AI sits inside production workflows.

Outage trackers such as Downdetector can show user-reported spikes, but they do not replace official incident reports. User reports can help show impact, while provider status pages confirm what the service operator has acknowledged.

The Bigger Reliability Question

The June 22 incident shows that AI reliability now matters to both consumers and businesses. As Claude expands into coding tools, APIs, and agentic work products such as Claude Cowork, service disruptions can ripple through more parts of the workplace.

For companies using Claude in critical workflows, the practical lesson is clear. AI providers should be monitored like any other cloud dependency, and teams should plan for retries, fallback models, cached responses, and graceful degradation.

Claude returned after the 90-minute multi-model incident, but the broader trend is harder to ignore. As demand for advanced AI models grows, reliability planning will become just as important as model quality.

FAQ