Verified September 2, 2026
Claude Fable 5.1 on OpenRouter
One endpoint, one key, forty providers — but OpenRouter sits between you and Anthropic, which changes pricing, caching, and what happens when the primary route fails.
The short answer
Supported. Call anthropic/claude-fable-5-1 through the OpenRouter chat/completions endpoint with your OpenRouter API key. Pricing mirrors Anthropic list rates plus OpenRouter's platform fee on credit purchases — check the live rate card before you commit. Prompt caching passthrough works when you send cache_control markers; routing fallbacks are your responsibility, not OpenRouter's default.
Support status
Available, verified September 2, 2026. OpenRouter lists anthropic/claude-fable-5-1 in its model catalog with Anthropic as the upstream provider. The model supports effort via provider-specific parameters and rejects temperature. OpenRouter does not remove Anthropic's 30-day retention requirement — data still flows through Anthropic's API regardless of which front door you use.
Setup
Four steps — step three is where production apps break.
- 1
Create an OpenRouter account and key
Sign up at openrouter.ai, add credits, and generate an API key. OpenRouter bills in USD credits; the rate card shows the per-token price including any platform markup at purchase time.
- 2
Use the namespaced model ID
Pass model: anthropic/claude-fable-5-1 in your request body. The anthropic/ prefix routes to Anthropic's upstream. A bare claude-fable-5-1 may resolve incorrectly or 404 depending on your client library.
- 3
Set effort, omit temperature
Pass effort through OpenRouter's provider routing parameters (see their Anthropic provider docs). Do not send temperature, top_p, or top_k — Fable 5.1 returns 400 invalid_request_error. Most starter templates set temperature by default; audit your SDK wrapper.
- 4
Handle refusals in your client
OpenRouter forwards Anthropic responses faithfully — a safety refusal arrives as HTTP 200 with stop_reason refusal, not as an thrown error. Branch on stop_reason and configure your own fallback model in the request or in client middleware.
Gotchas
The pricing line item nobody reads until the invoice arrives.
Credits include a platform markup
OpenRouter's displayed per-token rate can differ from Anthropic direct pricing because credits are purchased with a platform fee baked in. Compare the live OpenRouter rate card against platform.claude.com before you route production traffic — the gap matters at Agent-loop volume.
OpenRouter does not auto-failover on refusal
Unlike some gateways, OpenRouter forwards refusals as successful responses. If you need automatic retry on another model, configure models array or client-side fallback yourself — do not assume OpenRouter will silently reroute.
Prompt caching requires explicit markers
Anthropic's 90% cache-read discount applies only when you send cache_control breakpoints in the messages array. OpenRouter passthrough supports this, but generic OpenAI-compatible clients often omit cache markers entirely — you pay full input price on every turn.
30-day retention still applies
Routing through OpenRouter does not bypass Anthropic's mandatory 30-day retention on Fable 5.1. If your org requires zero-data-retention, Fable 5.1 is not available on any route — including OpenRouter.
Cost notes
List rates for Fable 5.1 are $10/$50 per million input/output (September 2026). OpenRouter adds a credit-purchase markup — check their rate card for the all-in figure. Prompt caching can cut effective input cost by 90% on cache hits if you implement cache_control correctly. Agent loops that omit caching and resend full repo context every turn dominate the bill regardless of gateway.
Compare gateway vs direct pricing
Run your token mix through the cost calculator with both direct Anthropic and gateway markup assumptions before you pick a route.