claude-opus-4-8 — pricing, context window and limits

The fallback target, not just another point release

Opus 4.8 shares Opus 5's price and request surface, but with one meaningful behavioural difference: thinking is off by default here, and only runs if you explicitly ask for it — the opposite of Opus 5's default. It's also the documented fallback destination when a request to Opus 5 or Fable 5 is refused, which makes it worth understanding on its own terms even if you don't choose it directly, since a fallback chain can route here without anyone deliberately selecting it.

What we could not confirm

Fast mode is supported on this model, but we could not confirm its specific fast-mode per-token rate, so no fast-mode figure is published for it here — if latency matters enough that fast mode is a real consideration, confirm the current rate directly before budgeting against it.

Why the default-off thinking behaviour matters for a fallback specifically

Because this model can be reached silently through a fallback path rather than a deliberate choice, code that assumes thinking runs by default — because that's how Opus 5 behaves — can be surprised the one time a fallback actually routes here and thinking doesn't run unless explicitly requested. Any fallback-aware code path should account for this difference explicitly rather than assuming uniform default behaviour across every model it might land on.

Related

See migrating: Opus 4.8 to Opus 5 for the direct diff against the model this one most often gets compared to.