June 2026 AI Tools Outage & Uptime Report
We continuously monitor 18 major AI tools including ChatGPT, Claude and Gemini. A recap of incidents recorded on providers’ official status pages in June 2026 (our first month of monitoring).
Longest outage: Anthropic（422h 9m）
By provider
| Provider | Incidents | Downtime | Worst state | Uptime |
|---|---|---|---|---|
| Anthropic | 10 | 443h 48m | Major outage | 38.36% |
| GitHub | 7 | 6h 53m | Major outage | 99.04% |
| OpenAI | 4 | 10h 30m | Partial outage | 98.54% |
| Cursor | 4 | 3h 21m | Major outage | 99.53% |
| ElevenLabs | 2 | 2h 10m | Partial outage | 99.7% |
| Replicate | 1 | 1h 7m | Partial outage | 99.84% |
| Runway | 1 | 19m | Partial outage | 99.96% |
Incident counts and downtime cover only "major-or-worse and resolved" incidents recorded on official status pages (minor latency and unresolved ongoing items are excluded). Uptime is our own estimate based on that, not official vendor figures. Aggregated by calendar month in Japan time (JST).
Highlights of the month
Early June saw frequent short-lived errors across chat and coding assistants. Anthropic (Claude) in particular logged repeated elevated errors on individual models such as Opus 4.7 and Sonnet 4.5/4.6, and on June 3 an incident affecting Claude Code. Most recovered within about an hour, though some elevated errors hit multiple models at once.
OpenAI saw "elevated error rates on Codex, ChatGPT and the Responses API" persist for roughly six hours on June 3 — the longest-impact incident of the month — alongside reduced ChatGPT availability for free users and ChatGPT Pro errors.
In coding tools, Cursor logged a brief but major degradation of Agent Sessions. GitHub recorded several disruptions including Code Scanning/Billing delays.
Generative media saw smaller blips too: ElevenLabs had transient STT/STS errors, Runway hit issues with Veo 3.1 and Seedance/Seedream generations, and Replicate saw contention on some GPU clusters (L40S/H200).
Trends & takeaways
Incidents clustered around conversational and coding services — Anthropic, OpenAI and Cursor — all heavily used, fast-moving areas. Yet most incidents resolved within about an hour and average uptime stayed high. Even when something "felt down," it was usually a brief, transient issue.
Outages were often announced per model (Opus, Sonnet, etc.). Since a single model can be degraded on its own, switching to another model or an alternative service is an effective workaround while you wait for recovery.
How to read this report
Incident counts and downtime include only incidents on each provider’s official status page that were both "major or worse (effectively unusable)" and already resolved (calendar month, Japan time). Minor degradations such as latency, and unresolved ongoing items (e.g. a deliberate suspension of a specific model), are excluded from the totals so the figures aren’t inflated beyond real impact.
Uptime is our own estimate based on that major-outage time — not official SLA figures.
Each service’s "outage history" page lists individual incidents with title, severity, duration and the official link.