Claude Opus 4.6 FAQ: Fixing failed switching, slowdowns, and off-spec outputs

When using Claude Opus 4.6, the most common frustrations aren’t “not knowing how to use it,” but rather not being able to find the entry point, replies suddenly becoming slow, or outputs not following requirements. Below is a consolidated Claude Opus 4.6 FAQ of high-frequency issues, organized by symptoms with an actionable troubleshooting order. Follow these steps and you can usually return to a stable, usable state within minutes.

Can’t find the entry point or switch to Claude Opus 4.6

If you can’t see Claude Opus 4.6 in the model selector, first confirm whether the current account has permission to use that model; the model list shown for the same account may not be exactly the same across different regions or different workspaces. Next, refresh the page and re-enter the chat page—sometimes Claude Opus 4.6 doesn’t appear simply because the frontend cache hasn’t updated.

If you still can’t switch to Claude Opus 4.6, try starting a new blank chat and test again; some older conversations “inherit” previous model settings, and switching can behave inconsistently. Finally, check whether browser extensions (scripts, ad blockers) are blocking the model menu from loading. Temporarily disabling the relevant extensions and then opening Claude Opus 4.6 usually fixes it.

Replies are very slow, stuck, or you see “system busy”

Claude Opus 4.6 may become slower or briefly queue during peak hours; this is a normal effect of service load. You can first split a one-time input into two parts: send “goal + format” first, then follow with “materials + constraints.” This often triggers more stable output faster than stuffing everything into a single long message.

If it keeps getting stuck, prioritize cleaning up your input: remove irrelevant quotes, long links, and repeatedly pasted text, then resend. Another effective approach is to reduce the “scope” of a single task—for example, change “write a complete plan” to “first list the outline and key decision points,” letting Claude Opus 4.6 run smoothly first and then refine step by step.

Output goes off-topic, ignores the format, or “over-embellishes”

Once Claude Opus 4.6 receives a vague goal, it may fill in assumptions on its own, making it look like it “wrote a lot but got it wrong.” The fix is to write your requirements in three parts: who you are (role), what you want (deliverable), and how it will be accepted (format and prohibited items). Add a final sentence like “If information is missing, ask questions first before continuing,” which can significantly reduce drift.

When you need a strict format (such as tables, JSON, or bullet lists), put a minimal example at the beginning and explicitly state “Only output the final result; do not analyze the process.” If Claude Opus 4.6 still won’t comply, in the next message send only “Reformat according to the previous format; do not change the content.” This is less effort than rewriting the entire instruction.

How to handle file/image upload failures and privacy concerns

If Claude Opus 4.6 encounters an upload failure, first rename the file in English and shorten the filename, then convert the file to a more common format (e.g., convert PDF to images, or save documents as a newer version). This can eliminate many compatibility issues. If image recognition is unstable, crop the image to keep only the key information area before having Claude Opus 4.6 interpret it—accuracy is usually higher.

When privacy is involved, it’s not recommended to dump personally identifiable information into Claude Opus 4.6 as-is; a safer approach is to anonymize first: replace names, phone numbers, and ID numbers with placeholders, and keep addresses only to the city level. You can also list sensitive fields separately as a “do-not-output list,” so Claude Opus 4.6 will automatically avoid them when organizing and summarizing.