Claude's Official Uptime Is 99.55%, But 45 Incidents in June — June AI Service Reliability Report

After monitoring 41 AI services (LLM APIs, coding agents, infrastructure, etc.) throughout June 2026, a clear polarization in reliability emerged between services.

Key Metrics Summary

• Highest reliability: Windsurf (0 incidents, 99.95% uptime) and Modal (94 points) showed the most stable performance.
• Lowest reliability: Deepgram (45 points) was the most unstable, recording 6 incidents totaling over 45 hours of downtime.
• Notable changes: GitHub Copilot showed improved performance with a significant score jump from 69 to 86 points, while Anthropic's services remained at a 'Fair' grade of 66-69 points.

Key Insights

• Decoupling of incident patterns: Incident count and incident duration (recovery speed) moved on different axes. For example, Codex had few incidents but long durations when they did occur, lowering its score, while Together AI had a very high incident count but defended its score thanks to fast recovery.
• Limitations of status pages: AIWatch's direct measurements captured multiple cases of latency degradation at services like Mistral and Replicate that were not reported on their official status pages. This suggests that even when a service isn't fully down, slower response times can significantly affect the actual user experience.