Anthropic reported elevated errors on Claude Opus 5 via its status page. The issue surfaced on Hacker News where the thread reached 96 points and 70 comments within hours.
Users described intermittent failures on both API calls and claude.ai, with some tasks timing out or returning empty responses. The incident page listed the event as ongoing at the time of the HN post.
What Happened
The status page recorded higher-than-normal error rates specifically for the Opus 5 model. No root cause was published in the initial update. Affected endpoints included both chat completions and long-context requests.
Early comments on the thread noted the problem appeared around peak usage hours in the US and Europe. Several developers reported the same prompt succeeding on one retry and failing on the next.
Scale of the Outage
The HN discussion logged 70 comments in the first day. Common reports included:
- 5xx errors on 15-30% of requests
- Increased latency above 8 seconds for successful calls
- Complete failure on 128k context windows
No official error-rate percentage was released by Anthropic at the time.
How to Check Status and Switch Models
Visit the live status page at https://status.claude.com for real-time updates. The incident link https://status.claude.com/incidents/mfdtrknpxghq shows the current state and any posted updates.
To maintain workflow continuity:
- Switch to Claude 3.5 Sonnet in the API or on claude.ai
- Update model parameter from
claude-opus-5toclaude-3-5-sonnet-20241022 - Monitor the status page before large batch jobs
Alternatives and Direct Comparison
When Opus 5 is unstable, teams typically fall back to other Anthropic models or competitors.
| Model | Context | Typical Speed | Error Rate (recent) | Best For |
|---|---|---|---|---|
| Claude 3.5 Sonnet | 200k | 1.2s | Low | General coding & writing |
| Claude Opus 5 | 200k | 2.8s | Elevated (incident) | Complex reasoning |
| GPT-4o | 128k | 0.9s | Stable | Speed-critical tasks |
Sonnet currently offers the closest capability profile with fewer reported interruptions.
Who Should Switch During Incidents
Developers running production agents or time-sensitive pipelines should route traffic away from Opus 5 until the incident resolves. Researchers needing maximum reasoning depth can keep Opus 5 for non-urgent experiments and queue jobs for later.
Teams already on Sonnet or Haiku saw no impact and can continue without changes.
Bottom Line
The outage highlights the risk of depending on a single high-capacity model. Routing logic that automatically falls back to Sonnet keeps most workflows running while Anthropic restores Opus 5 stability.
Anthropic has not yet published a post-incident report.
Top comments (0)