claude-opus-5 — pricing, context window and limits
| Status | current |
|---|---|
| Input / MTok | $5 |
| Output / MTok | $25 |
| Context window | 1,000K tokens |
| Max output | 128K tokens |
| Min cacheable prefix | 512 tokens |
| Effort levels | low, medium, high, xhigh, max |
| Thinking | adaptive-default |
| Platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
The default choice for agentic and enterprise work
Opus 5 is positioned as the workhorse of the current top tier — priced well below Fable 5 while covering the bulk of what enterprise and agentic coding workloads actually need. It's the model this site's own tools and reference tables treat as the practical default recommendation for demanding work that doesn't specifically require Fable 5's ceiling.
Two migration traps worth knowing before you switch
Thinking is on by default here, which is a change from several prior Opus generations, and it cannot be disabled once you're above the mid-to-high effort range — a request combining a high effort setting with thinking explicitly turned off is rejected. See thinking disabled above high effort on Opus 5 for exactly how this shows up. Separately, its minimum cacheable prefix is lower than several prior Opus generations, which is a genuine upside easy to miss: prompts too short to cache on an older Opus generation may cache cleanly here with no code change.
Fast mode is available, at its own separate rate
Where latency matters more than standard-speed pricing, Opus 5 supports a fast-mode tier at its own premium rate — see how fast mode changes the price of Opus for when that premium is actually worth paying.
Related
See migrating: Opus 4.6 to Opus 5 and migrating: Opus 4.8 to Opus 5 depending on which generation you're moving from.
Verified 2026-08-08 against claude-api skill — shared/models.md + SKILL.md model table (canonical: https://platform.claude.com/docs/en/about-claude/models/overview, https://platform.claude.com/docs/en/pricing).