Pricing · Fable 5 · Mythos · July 2026
Claude Fable 5 Pricing for Canadian Teams
Search demand for Fable prices is not academic. Claude Fable 5 is Anthropic’s most capable widely released model, and the sticker is high enough that Alberta owners feel it on the first big agent run. This page is the cost map in plain numbers.
Base API pricing (USD)
| Item | Fable 5 | Opus 5 (compare) |
|---|---|---|
| Base input | $10 / MTok | $5 / MTok |
| Output | $50 / MTok | $25 / MTok |
| 5m cache write | $12.50 / MTok | $6.25 / MTok |
| 1h cache write | $20 / MTok | $10 / MTok |
| Cache hit / refresh | $1 / MTok | $0.50 / MTok |
| Batch input / output | $5 / $25 | $2.50 / $12.50 |
Claude Mythos 5 (Project Glasswing, limited availability) is published at the same $10 / $50 base rates as Fable 5. Most Canadian SMBs will only ever touch Fable, not Mythos.
Worked cost examples (before tax / FX)
Input: 0.2 × $10 = $2.00
Output: 0.02 × $50 = $1.00
Total ≈ $3.00 USD for one heavy turn
Input: 0.2 × $5 = $1.00
Output: 0.02 × $25 = $0.50
Total ≈ $1.50 USD (~half of Fable)
Uncached 50k × $10 = $0.50
Cache read 450k × $1 = $0.45
+ output as usual
Caching is how Fable becomes usable for repeated agent loops
Multiply by ~1.35–1.40 for a rough CAD mental model only. Use your real FX and cloud invoice.
Other cost knobs that matter
- US-only inference: 1.1× multiplier on Claude 4.6+ when you force US residency.
- Tokenizer note: Fable 5 uses the newer tokenizer family. The same English text can produce roughly ~30% more tokens than older Claude generations. Budget with real logs, not old Opus 3 math.
- Safety fallbacks: Some cyber/bio queries route to lower models. Anthropic states you are not charged Fable rates for rerouted requests when fallbacks fire.
- 30-day retention: Fable use requires 30-day data retention for safety monitoring. Factor that into client contracts and municipal policy.
- Managed Agents: session runtime is billed separately ($0.08 per session-hour in Anthropic’s published table) on top of tokens when you use Managed Agents.
Consumer / claude.ai side (not just API)
Fable 5 is available to Pro, Max, Team, and Enterprise users, with plan rules that have changed since launch (June 9, 2026 launch, access disruption mid-June, restore July 1, then tighter paid access for heavy Fable use). Reporting in July 2026 described usage-based fees for Fable on consumer tiers at the same $10 / $50 rates as the API. Treat plan inclusions as volatile. Check Anthropic’s current Fable promotional / access article before you promise unlimited Fable to staff on a $20 plan.
When Fable is worth the 2×
- Multi-day coding migrations that already failed on Opus
- Hard research packs where one good pass beats three cheap loops
- Document-heavy finance / legal analysis with nested tables and charts
- Agent harnesses that run for hours with self-checks
When it is not worth it: daily email, simple CRM notes, high-volume classification. Use Sonnet 5 (intro $2 / $10 through August 31, 2026) or Haiku 4.5 for those.
Canada operator checklist
- Set monthly Fable budget caps in the Claude Console / cloud billing alerts.
- Default new apps to Opus 5 or Sonnet 5; escalate to Fable by flag, not by habit.
- Cache system prompts and tool schemas.
- Batch overnight jobs.
- Train staff so they stop retrying 10 times when one good brief would do. Bad prompting multiplies Fable burn.
Need a model routing plan for your team? Free call. Or bring a friend and claim a free training hour.
Book free call Bring a friend free hour Opus vs Fable guideRelated: Fable 5 launch overview · Opus 5 for Canada · Claude pricing hub