Is Claude Fable 5.1 more expensive than Fable 5?

No. Base token pricing is identical: $10 per million input tokens and $50 per million output tokens, unchanged from Fable 5. The change is in cache reads, which dropped 75% to $0.25 per million tokens — a 0.025x multiplier on input price, rather than the 0.1x multiplier every other Claude model uses. Anthropic estimates roughly 25% lower cost on typical cache-heavy workloads, so for agent loops that re-read a large prefix each turn, Fable 5.1 is meaningfully cheaper than Fable 5 at the same headline rate.

Should I use Fable 5.1 or Opus 5?

It depends on whether your work is price-sensitive or capability-sensitive. Opus 5 costs $5/$25 per million tokens — half Fable's input price and half its output price — and remains the sensible default for everyday coding and agent work. Fable 5.1 costs $10/$50 and, on Anthropic's launch benchmarks, leads Opus 5 on every reported evaluation, including 55.8% versus 52.3% on Terminal-Bench 4.0 and 52.6% versus 29.0% on Terminal-Bench-Science. The science gap is the widest and the clearest reason to pay the premium.

Do I need to change my code to use Fable 5.1?

You need to change the model identifier, and you should re-check any effort-level configuration. Fable 5.1 delivers similar or better results than Fable 5 at low and medium effort and pulls further ahead at higher effort tiers, so a prompt tuned to compensate for Fable 5's behaviour at a given effort level may now be over- or under-provisioned. Cache configuration needs no change to benefit from the cheaper reads — the new multiplier applies automatically.

Who can access Claude Mythos 5.1?

Only approved organisations. Fable 5.1 is generally available through the API and Claude apps, while Mythos 5.1 — the same underlying model with different safeguards — is limited to trusted access programmes for specific cybersecurity and life-sciences organisations. This mirrors the split Anthropic used for Fable 5 and Mythos 5 in June 2026.

Quick Answer

Claude Fable 5.1 vs Fable 5 vs Opus 5: What Changed

Published: September 2, 2026

The Short Answer

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. Three things actually changed, and one thing pointedly did not:

Price did not change. Fable 5.1 is still $10 per million input tokens and $50 per million output tokens.
Cache reads got 75% cheaper — $0.25 per million tokens, down from $1.00 on Fable 5.
Benchmarks moved up, with the largest jump on scientific reasoning.
False refusals went down, which is the quality-of-life change most daily users will notice first.

Last verified: September 2, 2026.

Side by Side

	Fable 5.1	Fable 5	Opus 5
Released	Sep 1, 2026	Jun 2026	Jul 24, 2026
Input / MTok	$10	$10	$5
Output / MTok	$50	$50	$25
Cache read / MTok	$0.25	$1.00	$0.50
Cache multiplier	0.025x	0.1x	0.1x
Terminal-Bench 4.0	55.8%	lower	52.3%
Terminal-Bench-Science	52.6%	lower	29.0%
Availability	GA	GA	GA
Best for	Hard science, long agent runs	— (superseded)	Default coding/agents

Benchmark figures are Anthropic’s own launch numbers, published September 1, 2026. Treat vendor benchmarks as a directional signal, not an independent result — the only number that matters for your decision is the one your own eval suite produces.

The Cache Read Change Is the Real Story

Every other Claude model prices a cache hit at 0.1x the input rate. Fable 5.1 and Mythos 5.1 price it at 0.025x. That is not a rounding adjustment; it is a different pricing philosophy for the same product tier.

Concretely, a cached prefix re-read costs:

$1.00 per MTok on Fable 5
$0.50 per MTok on Opus 5
$0.25 per MTok on Fable 5.1

So Fable 5.1 re-reads cached context at a quarter of Fable 5’s cost and half of Opus 5’s cost, despite carrying twice Opus 5’s headline input price.

This matters because of how agent workloads actually consume tokens. A coding agent working through a large repository re-sends a substantial system prompt, tool schema and file context on every single turn. In a 30-turn session, the cached prefix is read 30 times and written once. Under that shape, the cache-read multiplier dominates total spend far more than the headline input rate does — which is why Anthropic can leave $10/$50 untouched and still describe typical workloads as roughly 25% cheaper.

The practical rule: if your prompt-cache hit rate is high, Fable 5.1’s effective cost is much closer to Opus 5 than the price sheet suggests. If you are doing one-shot, low-cache calls, you get the benchmark gains and none of the savings, and Opus 5’s $5/$25 remains the better economics.

Where Fable 5.1 Actually Wins

Scientific reasoning is the standout. The Terminal-Bench-Science gap — 52.6% versus Opus 5’s 29.0% — is not an incremental improvement; it is a different capability tier on that evaluation. If your workload is computational biology, chemistry tooling, data analysis over experimental results, or anything that combines domain science with terminal execution, this is the strongest argument Anthropic has made for the Fable premium since the line launched.

Coding is a narrower win. 55.8% versus 52.3% on Terminal-Bench 4.0 is a real but modest 3.5-point gap for a 2x price difference. For general software work, that arithmetic favours Opus 5 unless your failure cases are concentrated in exactly the hard tail Fable is better at.

Effort scaling changed shape. Anthropic reports similar or better results than Fable 5 at low and medium effort, with much larger gains at higher effort tiers. The implication for cost control: on Fable 5.1 the high-effort tier buys you more than it used to, so if you previously settled on medium effort because high was not worth the tokens, that trade-off is worth re-testing.

Fewer false refusals. Fable 5 drew consistent complaints about over-refusal on legitimate security and life-sciences work. Anthropic explicitly targeted this in 5.1. Anyone who abandoned Fable for benign-but-flagged prompts should re-run those cases.

Should You Upgrade?

Upgrade now if: you run cache-heavy agent loops (the 75% cache cut is free money), you do scientific or research-heavy work, or you hit false refusals on Fable 5.

Stay on Opus 5 if: you are doing mainstream coding and agent work where 52.3% versus 55.8% does not justify paying 2x on both input and output.

Do not upgrade blindly if: you have prompts tuned around Fable 5’s effort-level behaviour. Re-run your evals before switching production traffic — the effort curve moved.

What This Says About the Frontier Price Order

As of September 2026 the pricing ladder is genuinely strange. Fable 5.1 sits at $10/$50, Opus 5 at $5/$25, and GPT-5.6 Sol at $4/$20 after its August 21 cut. The most expensive model in the Claude line is now more than twice the price of OpenAI’s flagship, while Anthropic’s own Opus 5 undercuts its premium tier at every token category except cache reads — where Fable 5.1 is suddenly the cheapest of the three.

Vendors are no longer competing on a single headline number. They are competing on the shape of the bill, and prompt caching is where that competition now happens.