Claude Fable 5.1 Pricing: $10/$50 Per Million Tokens, Plus a 75% Cache Cut

By Wallace Gomes de Souza (LW Forge) — about the author · Updated September 7, 2026

Anthropic shipped Claude Fable 5.1 on September 1, 2026, and the pricing tells a different story than the benchmark chart does: at $10 per million input tokens and $50 per million output tokens, it's not a cheaper Claude — it's a more expensive one, priced above every model already in Anthropic's lineup. The one number that cuts the other way is cache: reads drop 75%, which is where most of the savings Anthropic is advertising actually come from.

The actual numbers

Confirmed in Anthropic's own launch announcement:

That cache-read rate is a 75% cut from what it replaces — back-calculating from Anthropic's own percentage, the prior rate was roughly $1.00 per million. Anthropic reports that cut brings typical workload costs down about 25% and agentic-workload costs down up to 45%, since agentic pipelines re-send the same tool definitions and context on nearly every call, which is exactly where cache reads pile up.

Where Fable 5.1 sits against the rest of the Claude lineup

Here's the part the benchmark coverage skips: Fable 5.1 is priced well above Anthropic's existing flagship.

Fable 5.1 costs exactly 2x Opus 5 on both input and output — and 5x Sonnet 5. Nothing else in Anthropic's public catalog sits at this price point. That positions Fable 5.1 as a specialized tier above the general-purpose lineup, not a drop-in upgrade for whatever is currently calling Opus or Sonnet. Its sibling, Claude Mythos 5.1, shares the same underlying weights but runs a different safeguard configuration and is currently gated to vetted organizations through Anthropic's Cyber Verification and Life Sciences Verification programs — it isn't something you can price into a general product today.

What the capability jump actually looks like

The price increase tracks a real capability jump, not just a rebrand. Per Anthropic's benchmarks, Fable 5.1 moves past Fable 5 by a wide margin:

• Agentic coding: 42.0% → 55.8–60.9%
• Agentic scientific research: 24.7% → 52.6%
• Computer use (OSWorld 2.0, partial credit): 72.9% → 77.9%

That's the case for the higher price — Fable 5.1 is being positioned for agentic and research workloads that specifically need the ceiling, not for the volume of calls a typical product sends.

What it costs in practice

Take a coding agent that runs 150 sessions a day. Each session sends roughly 3,000 tokens of fresh input, 25,000 tokens of cached context (system prompt, tool definitions, prior turns re-sent on every step — the normal shape of an agentic loop), and returns 1,200 tokens of output. Over a 30-day month:

• Fresh input: 3,000 × 150 × 30 = 13.5M tokens → 13.5 × $10.00 = $135
• Cached input: 25,000 × 150 × 30 = 112.5M tokens → 112.5 × $0.25 = $28.13
• Output: 1,200 × 150 × 30 = 5.4M tokens → 5.4 × $50.00 = $270
• Total: ≈ $433/month

Notice the shape: output dominates the bill even though cached input is by far the largest token count — at $50/M, output is 200x more expensive per token than a cache read. If a coding agent's response length grows, the bill grows with it far faster than the context it's reading grows. Run your own session shape through the Claude cost calculator — Fable 5.1 isn't in the model list there yet (this site's calculators price the general-availability lineup: Opus 5, Sonnet 5, Haiku 4.5), so use Opus 5's rate as the closest stand-in and multiply by 2x on input and output to approximate Fable 5.1 until the catalog is updated.

Should this change what you're calling today

For most products sending routine calls — chat, summarization, classification — this launch changes nothing: Opus 5, Sonnet 5 and Haiku 4.5 are unaffected, and Fable 5.1's price point puts it well outside a typical per-call budget. It matters specifically if you're building an agentic coding tool, a research pipeline, or anything else where the last 15 percentage points of accuracy are worth 2x Opus 5's rate. For everything else, the more relevant Anthropic story this quarter is still Sonnet 5's launch price becoming permanent — that's the one that actually moved the bill for most teams. If you're comparing where each Claude tier fits, Opus vs. Sonnet vs. Haiku has the full breakdown, and the token calculator turns any of these per-million rates into what your actual traffic would pay.

Frequently asked questions

What is Claude Fable 5.1?

Claude Fable 5.1 is Anthropic's September 2026 model release, priced at $10 per million input tokens and $50 per million output tokens, with large benchmark gains in agentic coding and scientific research over its predecessor, Fable 5.

How much does Claude Fable 5.1 cost per million tokens?

$10.00 for input, $50.00 for output, and $0.25 for cache reads — a 75% cut to cache pricing versus Fable 5, which Anthropic says lowers typical workload cost by about 25% and agentic-workload cost by up to 45%.

Is Claude Fable 5.1 cheaper than Claude Opus 5?

No — it's more expensive. Fable 5.1 costs exactly 2x Opus 5 on both input ($10 vs. $5) and output ($50 vs. $25), and 5x Claude Sonnet 5. It's priced as a specialized tier above the rest of Anthropic's public lineup, not a cheaper alternative to it.

What is the difference between Claude Fable 5.1 and Claude Mythos 5.1?

They share the same underlying model but run different safeguard configurations. Fable 5.1 is generally available on the API, AWS, Google Cloud, and Azure. Mythos 5.1 is currently restricted to vetted organizations through Anthropic's Cyber Verification and Life Sciences Verification programs.

Is Claude Fable 5.1 available in the calculators on this site?

Not yet — the calculators on this site currently price Anthropic's general-availability lineup (Opus 5, Sonnet 5, Haiku 4.5). Until Fable 5.1 is added, doubling Opus 5's input and output rate gives a close approximation.