Two facts about Claude Opus 5 capacity catch teams out in production, and both are structural rather than something you can prompt your way around. Check them before you route traffic, not after.
It has its own rate-limit bucket
Opus 4.8, 4.7, 4.6, and 4.5 share a single combined Opus limit. Claude Opus 5 does not draw on it.
The consequence runs both ways: shifting traffic from Opus 4.8 to Opus 5 does not free headroom on the old bucket in any way you benefit from, and Opus 5 does not inherit the limits you were already comfortable with. Check your tier's Opus 5 limits explicitly before moving volume.
Priority Tier does not cover it
Priority Tier is supported on every other current model, including Claude Fable 5 and Opus 4.8 — but not on Claude Opus 5, Claude Sonnet 5, or Claude Mythos 5. A Priority Tier request naming one of these fails validation.
If a latency commitment is contractual for you, that is a genuine reason to keep a workload on Opus 4.8.
Handling limits gracefully
- Respect
retry-after. A 429 response includes it; the official SDKs retry automatically with backoff. - Fast mode has a separate pool, so it can absorb overflow — at double the price.
- Batch anything that can wait. The Batch API is 50% cheaper and takes pressure off interactive capacity.
- Cascade down, not up. On sustained pressure, route overflow to Sonnet 5 rather than queueing everything behind Opus 5.
More on Claude Opus 5
Start with the complete Claude Opus 5 guide for the overview, or go deeper:
Ready to go deeper?
Read the full Claude Opus 5 guideFrequently Asked Questions
Does Claude Opus 5 share rate limits with Opus 4.8?
No. Opus 4.8, 4.7, 4.6, and 4.5 share a combined pool; Claude Opus 5 has its own. Moving traffic between them neither frees headroom on the old pool nor inherits its limits, so check your tier's Opus 5 limits before shifting production volume.
Is Claude Opus 5 available on Priority Tier?
No. Priority Tier covers most current models including Claude Fable 5 and Opus 4.8, but excludes Claude Opus 5, Claude Sonnet 5, and Claude Mythos 5 — a Priority Tier request naming one of them fails validation.
What should I do when I hit a rate limit?
Respect the `retry-after` header — the official SDKs handle this automatically with backoff. For sustained pressure, move deferrable work to the Batch API at half price, route overflow to a cheaper model, or use fast mode, which draws on a separate pool at double the cost.
