Claude Fable 5 pricing, context window, and cost calculator
Anthropic's premium generally available model for demanding reasoning and long-horizon agentic work.
Verified fields
Quick facts
claude-fable-5Supported capabilities
- Streaming
- Prompt caching
- Reasoning
Only capabilities explicitly tracked from provider documentation are shown.
Source-tracked rates
Verified pricing profiles
Amounts are USD per 1M tokens unless the column states an hourly storage unit.
| Profile | Input / 1M | Cached input / 1M | Cache write / 1M | Output / 1M | Notes |
|---|---|---|---|---|---|
| standardDefault | $10.00 USD | $1.00 USD | $12.50 USD (5m) / $20.00 USD (1h) | $50.00 USD | Used for default estimates. |
Workload estimate
Claude Fable 5 cost calculator
Change the workload assumptions. The calculator uses the same shared pricing resolver as StackLens Compare.
Estimated monthly total$350.00
- Monthly input cost
- $100.00
- Monthly output cost
- $250.00
- Per 1,000 requests
- $35.00
- Active profile
- standard
Estimate excludes untracked provider-specific charges.
Explicit assumptions
Common workload examples
1M input tokens
$10.00 under the standard profile, excluding output and other charges.
1M output tokens
$50.00 under the standard profile, excluding input and other charges.
10,000 monthly requests
At 1,000 input and 500 output tokens per request: $350.00 under standard.
50% cached input
The same default workload with 50% cached input is estimated at $305.00. Cache savings apply only to the tracked cached-input rate.
Research layer
Experience and rollout notes
Official facts are separated from user reports. Community reports are useful signals, not controlled benchmarks.
Official positioning and limits
Anthropic positions Claude Fable 5 for demanding reasoning and long-horizon agentic work. Its official documentation lists a 1M-token context window, up to 128K output tokens, and separate refusal and fallback behavior.
What this does not prove: This describes provider-documented behavior, not an independent quality result.
Long-running coding work
Early users repeatedly describe Fable 5 as more independent on complex coding tasks and as needing fewer review cycles in some long-running workflows.
What this does not prove: These are self-selected early reports. Task difficulty, tooling, prompts, and plan limits were not controlled.
Refusals, fallbacks, and usage
Multiple early reports flag classifier refusals or fallback interruptions during coding work, alongside concern about premium usage and token cost. Other users report no comparable disruption.
What this does not prove: Product-plan behavior and API behavior may differ. Treat this as a rollout risk to test, not a measured failure rate.
No controlled StackLens benchmark yet
StackLens has not run a controlled Fable 5 benchmark. The notes above separate official documentation from independent user reports.
Research reviewed 2026-07-13. Reports may change as Claude Fable 5 reaches more workflows.
Tracked relationships
Related models
StackLens relationship map based on model families and published decision comparisons. Find a workload match → Open relationship map → Open price map →
Featured comparisons
- Claude Fable 5 vs Kimi K3 cost
- Claude Fable 5 vs Grok 4.5 cost
- Claude Fable 5 vs GPT-5.6 Sol cost
- Claude Fable 5 vs DeepSeek V4 Pro cost
- Claude Fable 5 vs Claude Sonnet 5 cost
- Claude Fable 5 vs Claude Sonnet 4.6 cost
Continue your research
Verification
Sources and methodology
Pricing and limits are source-tracked and may change. Verify current values with the provider before making production purchasing decisions.
Calculator profile rule: standard is the verified default. Claude Fable 5 can return classifier refusals as successful HTTP 200 responses with stop_reason set to refusal. Anthropic documents fallback handling for these responses.