llmrelay
// claude

Claude Fable 5.1

Anthropic’s current top Claude. Same $10/$50 list, cheaper cache reads.

Claude Fable 5.1 shipped 1 September 2026. Anthropic’s own page calls it and Mythos 5.1 the most advanced models they sell for coding and knowledge work — not a creative-writing sideline. Mythos 5.1 is the same weights with looser safeguards, invitation-only. The public id is claude-fable-5-1 (hyphen). claude-fable-5.1 returns no channel.

List price is unchanged from Fable 5: $10 / $50 per million tokens. Cache hits dropped to $0.25/M from $1/M. Anthropic says that cuts typical billed work about 25%, and highly agentic work up to about 45%. We sell input and output at half list — $5.00 / $25.00. We do not currently pass CacheRatio through, so the cheaper cache is our margin, not a line on your invoice.

Anthropic still tells you to start on Opus 5 and reach for Fable 5.1 when Opus evals fall short. Adaptive thinking is always on. Claude Code defaults to high effort; Cowork and claude.ai default to medium. Reddit on launch day was demos (Cities: Skylines in three.js, a Minecraft mod) and a warning that the wrong effort setting burns tokens.

Official list
llmrelay
Input / M tokens
$10.00
$5.00
Output / M tokens
$50.00
$25.00
Context window
—
1,000,000 tokens
Max output
—
128,000 tokens

What it costs you per month

Real-world budget scenarios. Numbers are simple sums — official list price vs llmrelay's 50% off tier.

Usage scenario
Official
llmrelay
Light coding (1M in / 200K out per month)
$20.00
$10.00save $10.00
Heavy Cursor / Cline user (50M in / 5M out per month)
$750.00
$375.00save $375.00
Production RAG (500M in / 20M out per month)
$6000.00
$3000.00save $3000.00

What it's good at

Best for

Pick Claude Fable 5.1 when

  • +Opus 5 already failed the specific task. That is Anthropic’s own trigger, and the only honest one at 2× the Opus rate.
  • +You are in Claude Code on a hard, long-horizon job: whole-repo change, root-cause review, multi-hour agent loop.
  • +You were on Fable 5 and want the current snapshot. Same $5/$25 here, newer cutoff (June 2026 on Anthropic’s table), cheaper cache on their card.
  • +You want one key that also runs gpt-6-astra, so you can A/B the two flagships without a second vendor.

Choose something else when

  • !Everyday coding. Opus 5 is $2.50/M input here. Anthropic rates Fable slower, and thinking cannot be turned off, so easy prompts still pay for reasoning tokens.
  • !High-volume loops. At $5/M input the bill grows faster than the 2× rate card because always-on thinking emits more tokens per task.
  • !You need a native image or video model. Fable writes code and documents; it does not emit pixels.
  • !You typed claude-fable-5.1 with a dot. That id 503s. Use claude-fable-5-1.

Questions people ask about Claude Fable 5.1

How can I use Claude Fable 5.1?

Officially: Claude.ai, Claude Code, the Claude API as claude-fable-5-1, plus AWS, Google Cloud and Azure. On llmrelay, set base URL to https://api.llmrelay.dev/v1 and model to claude-fable-5-1. Hyphen, not a dot. Same prepaid key as Opus 5 and GPT-6 Astra.

How much does Claude Fable 5.1 cost?

Anthropic list is $10.00 per million input and $50.00 per million output — same as Fable 5. Cache hits are $0.25/M, a quarter of Fable 5. On llmrelay it is $5.00 and $25.00, half list. We do not currently pass the cache-hit discount through to the bill.

Is Fable 5.1 a writing model?

No. Anthropic’s 5.1 launch page is coding, knowledge work, computer use and research. The older “narrative specialist” framing was Fable 5 marketing we had on this site; 5.1 is the escalation above Opus 5, at twice the token rate.

Should I use Fable 5.1 or Opus 5?

Default to Opus 5 at $2.50/$12.50. Reach for Fable 5.1 at $5.00/$25.00 when Opus 5 has already failed that task, or when you are paying for a once-off migration where a wrong answer costs more than the tokens. One key covers both.

Should I use Fable 5.1 or GPT-6 Astra?

Same $5.00/$25.00 on llmrelay. Fable if you are in Claude Code and already know Opus 5 is not enough. Astra if you need OpenAI computer use, Codex, or ChatGPT Sites. Run the same prompt on both and keep the one that finishes the job.

What is the context window on Claude Fable 5.1?

1,000,000 tokens input, with up to 128,000 output tokens. Anthropic’s table lists a June 2026 knowledge cutoff and latency rating Slower. Adaptive thinking is always on.

Try Claude Fable 5.1 at half the price

Free to create an account, no subscription, $10 minimum top-up — enough to run this model against a real task and judge quality yourself.

Get API key →

Compare Claude Fable 5.1 against the alternatives

The comparisons this model appears in, the models nearest it on price, and the full rate card.