Claude Fable 5.1 API Pricing

Anthropic's pricing table lists $10/M base input, $12.50/M 5-minute cache writes, $20/M 1-hour cache writes, $0.25/M cache hits, and $50/M output. Batch is half of base at $5/M and $25/M. Under this site's 92/8 agentic blend at an 82% cache-hit rate the effective rate is $6.2586/M.

One number. Base input, both cache-write tiers, output and the batch rates are byte-for-byte identical to Fable 5. The cache hit went from $1/M to $0.25/M. Anthropic prices cache reads at 0.1x base input on every other Claude model; on Fable 5.1 and Mythos 5.1 it is 0.025x, and the pricing page carries a footnote saying precisely that.

Because the two figures measure different workloads, and both are correct. Our standard blend is 92% input / 8% output at an 82% cache-hit rate, and it charges every cache miss at the $12.50/M write rate. That leaves the effective rate at $6.2586 against Fable 5's $6.8244 - a 8.3% saving. Push the cache-hit rate to a theoretical 100% and our blend still only reaches 14%, because output is unchanged and contributes $4 of the total on its own. Anthropic's 45% therefore describes a much more input-heavy mix than 92/8, with many reads per write. If your agent re-reads one long cached context hundreds of times per write, you will land closer to the vendor's number than to ours.

When input dominates and the same cached prefix is read many times. The gap between a read ($0.25) and a 5-minute write ($12.50) is 50x on this model - the widest ratio on any Claude row - so what decides your bill is not the hit rate alone but how many reads each write earns. A long-running agent on one stable context is the best case; short unrelated requests that each write a fresh cache are the worst, and there Fable 5.1 costs the same as Fable 5.

Opus 5 is exactly half the card at $5/$25 and blends to $3.4122/M, so Fable 5.1 costs 1.8x as much on our assumptions. Anthropic's own guidance is to start with Opus 5 for most workloads and reach for Fable 5.1 for demanding reasoning and long-horizon agentic work, or when evals on Opus 5 at higher effort still fall short.

The same split as the 5.0 pair: identical rate cards, different access. Fable 5.1 is generally available with Anthropic's production safeguards. Mythos 5.1 is limited availability through the Glasswing programme for vetted cybersecurity and life-sciences organisations, with no self-serve API.

Not directly. Anthropic notes that Claude 4.7 and later models use a newer tokenizer producing roughly 30% more tokens for the same text, with the exact increase depending on content. A model on the newer tokenizer at the same nominal $/M is more expensive per page of real text than one on the older tokenizer. Anthropic puts 1M tokens at about 555k words on the current tokenizer against about 750k words before it.

No. Anthropic's pricing page states that Claude 4.6 and later models include the full 1M-token context window at standard pricing - a 900k-token request is billed at the same per-token rate as a 9k-token one - and that caching and batch discounts apply at standard rates across the whole window. Pinning inference to the US with inference_geo is the modifier that does cost more: a 1.1x multiplier on every token category.