Opus 5 vs Fable 5
Newer releases exist: Opus 5 → Opus 5.5; Fable 5 → Fable 5.1. See Opus 5.5 vs Fable 5.1 for the same choice between the current models.
Side by side
| Opus 5 | Fable 5 | |
|---|---|---|
| Input / MTok | $5 | $10 |
| Output / MTok | $25 | $50 |
| Cache read / MTok | $0.50 | $1 |
| Context window | 1M | 1M |
| Max output | 128K | 128K |
| Min cacheable prefix | 512 | 512 |
| Effort levels | low, medium, high, xhigh, max | low, medium, high, xhigh, max |
| Effort default | high | high |
| Thinking | On by default — can be disabled | Always on — cannot be disabled |
| Forced tool_choice | accepted | accepted |
The previous top two tiers, and what separates them
Fable 5 sits above Opus 5 in both price and positioning, with thinking that runs on every request and cannot be turned off at all, versus Opus 5's default-on-but-adjustable behaviour. For most workloads that would consider either, Opus 5's lower price and more flexible thinking control make it the more practical default, with Fable 5 reserved for the narrower set of tasks that genuinely need the highest ceiling available.
The constraint that decides it before price even enters the picture
Before weighing cost, check data-retention requirements: Fable 5 requires a retention window incompatible with a zero-data-retention organisational policy, while Opus 5 has no such restriction. For an organisation under that policy, this isn't a cost-versus-quality trade at all — Fable 5 simply isn't an eligible option regardless of how the rest of the comparison looks.
When the extra cost is actually justified
Reach for Fable 5 specifically when a task has already been tried against Opus 5 and fallen short in a way that matters — not as a default first choice for anything that merely sounds important. The price gap between them is real enough that it should be earned by a demonstrated need, not assumed from a task's perceived seriousness alone.
Related
See Fable 5 is unavailable to zero-data-retention organisations for exactly how that constraint applies, and the model picker for a recommendation that accounts for it automatically.
Verified 2026-09-30 against claude-api skill — shared/models.md, model-migration.md + SKILL.md model table and Anthropic's pricing and model-deprecation pages (canonical: https://platform.claude.com/docs/en/about-claude/models/overview, https://platform.claude.com/docs/en/about-claude/pricing, https://platform.claude.com/docs/en/about-claude/model-deprecations).Could not confirm: Availability on Amazon Bedrock, Google Cloud and Microsoft Foundry is not stated in the source we used, so we do not claim it either way.