When Claude Opus 5 suddenly starts giving flat or oddly off answers mid-conversation, that "did it just get worse?" feeling usually traces back to one of three documented causes: a safety classifier silently switching your conversation to a lighter model, a temporary service incident, or hallucination — a known limitation where the model generates plausible-sounding but false content. Here's how to tell them apart.
🟢 Claude Opus 5 · model notice · Fable subscription
Fable 5 and 5.1 subscription (updated September 7, 2026): Claude Fable 5.1, released September 1, 2026, is the current Fable model and Fable 5 is now legacy. Plan terms are the same for both — Max and Team Premium plans include Fable at up to 50% of the weekly usage limit; Pro and Team Standard use usage credits (
Why does Opus 5 suddenly feel like a different model?
The most common cause is automatic model switching triggered by a safety classifier. According to official support documentation, Opus 5 has classifiers that flag higher-risk cybersecurity requests — such as exploit generation, binary-based vulnerability scanning, and penetration testing — and this review covers all input sources, not just your typed message, including memory, connectors, web search, and files. When a request is flagged, Claude automatically re-runs it on Opus 4.8, described as "our next-most-capable model," within the same conversation. A notice shows which model answered, and the conversation stays on that model for the rest of the chat unless you switch back manually. If replies suddenly feel blunter or less sharp, a lighter model may genuinely be the one answering.
Cause 1: Automatic model switching — what should you check on screen?
First, check the model label shown with the response. If a conversation started on Opus 5 but at some point shows Opus 4.8 as the responder, your previous request was flagged by the safety classifier and auto-switched. The flow below shows how this works.
If you'd rather not have this happen automatically, look in Settings under Capabilities for a toggle related to "switch models when a message is flagged." Turning it off pauses the conversation for your input instead of auto-switching. The exact label can change between updates, so confirm the current wording on screen.
Cause 2: Temporary service incidents — how often, and how do you check?
If Claude's status page shows an actual error-rate spike during the time you noticed odd answers, that explains it. Recently, on August 19, 2026, Opus 5 and Haiku 4.5 saw elevated request errors that were identified and fixed, and on August 24, 2026, several models including Opus 5 saw elevated errors for a few hours before recovering. During windows like these, responses can be slower, cut off, or lower quality than usual.
- How to check — visit status.claude.com and see whether the Opus 5 component was marked as something other than "Operational" during that time.
- What to do — incidents typically resolve on their own, so retry the same request a bit later.
- Caveat — if odd answers keep recurring with no incident on record, suspect model switching or hallucination first instead.
Cause 3: Hallucination — what do you do when a confident answer is just wrong?
Hallucination — when an AI model generates plausible-sounding content that isn't actually grounded in fact — is described in official support documentation as "a byproduct of some of the current limitations of frontier Generative AI models." If a reply sounds unusually confident but turns out to be wrong, this may be why.
- Outdated training data — Claude can lack current information and get confused when asked about recent events.
- Fabricated quotes — it can produce quotes that look authoritative or sound convincing but aren't grounded in fact.
- Source quality — when web search is used, answer quality depends on the quality of the underlying sources it references.
Official guidance recommends not relying on Claude as a singular source of truth, scrutinizing high-stakes advice carefully, and checking the original sources it cites for context Claude may have missed. If an answer is clearly wrong, using the thumbs-down feedback button under the response is one way to report it.
Comparing the three causes at a glance
| Cause | Main symptom | How to check | What to do |
|---|---|---|---|
| Automatic model switching | Answer quality drops from a certain point on | Check the model label on the response | Turn off the auto-switch toggle in Settings if desired |
| Temporary service incident | Slow, dropped, or erroring responses | Check status.claude.com | Wait and retry later |
| Hallucination | Confident but factually wrong content | Cross-check quotes and cited sources | Verify against the original source, submit feedback |
Frequently Asked Questions
Q. I got a notice that Opus 5 switched to Opus 4.8 mid-conversation. Why?
Your previous request was likely flagged by the safety classifier as a high-risk cybersecurity request, such as exploit generation or penetration testing content. Claude then automatically re-runs it on Opus 4.8 within the same conversation, and stays on that model afterward unless you switch back.
Q. Can I turn off automatic model switching entirely?
Yes — in Settings under Capabilities, there's a toggle related to switching models when a message is flagged. Turning it off pauses the conversation for you to decide instead of auto-switching. Confirm the exact wording on screen, since it can change between updates.
Q. There's no switch notice and no incident on the status page, but answers still feel off. What should I suspect?
Hallucination is worth checking first. This is especially true for answers about recent events or specific quotes — Claude may have generated plausible but ungrounded content, so verify against the original source.
Q. How do I check status.claude.com for past incidents?
Visit the site and look at component status for the relevant time window. Recently, on August 19 and August 24, 2026, some models including Opus 5 saw temporary elevated error rates that recovered within a few hours.
Sources: Why Claude switched models in your conversation with Opus 5, Claude is providing incorrect or misleading responses, Claude Status.
Related articles
- Troubleshooting Claude Error Messages: Limits, Capacity, Login
- Claude Not Working? Fix Login, Access & Errors (2026)
- Claude File Upload FAQ: Formats, 500MB/20-file Limits, PDF Handling
- Fix "Your Previous Message Wasn't Sent" in Claude — Causes & Steps (2026)
- How to Check Claude Status — Is Claude Down Right Now?