“Claude Fable is too expensive” is the wrong sentence to be reading in September 2026. It was written about Fable 5 — the model Anthropic upgraded on September 1 with Claude Fable 5.1, which keeps the same $10 / $50 per-million-token rate card but cuts cache-read pricing by 75%, a change Anthropic says makes typical workloads about 25% cheaper and highly agentic ones up to 45% cheaper. The pricing guides that still rank for “claude fable price” were published in June and describe the previous model.
So here is the current answer, with every figure captured on September 4, 2026: Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens on Anthropic’s own pricing, with cache reads at $0.25 per million tokens (75% less than Fable 5). If you call it through Modellix, an API gateway that lists Fable 5.1 (and 27 other language models) behind one key, the same model bills at $9 input / $45 output / $0.225 cache read, with a Discount column that reads “10% OFF” — captured on the Modellix LLM price page the same day. We run Modellix, so we have a commercial interest in the gateway comparison below; where a number could not be verified against Anthropic’s own pages, we say so explicitly.
This is a pricing guide for one model ID, not a model review: it covers the four dimensions a Fable bill is built from, what the Fable 5 → 5.1 migration changes in dollar terms, what the same model costs through Modellix and how to reproduce that on the price page’s calculator, and how to audit the bill after the run. Whether Fable 5.1 is worth using instead of Opus 5 is a separate decision question — the Fable 5.1 vs Opus 5 guide takes that on, and the Claude API pricing guide covers Opus 5, Sonnet 5, and Haiku 4.5 for teams comparing the whole Anthropic lineup.
Claude Fable price: the current rate card (September 4, 2026)
All rates are USD per 1 million tokens — the only unit Anthropic bills in — and Fable 5.1 is flat-rate: one input price and one output price across its full 1M-token context window, with no surcharge for long contexts (the price table’s Input Context column shows “—“ for it, the flat-rate marker). The four rates, as listed on the Modellix LLM price page on September 4, 2026, with the page’s own “official list price → price you pay” formatting:
| Dimension | Official list price | On Modellix | Discount column |
|---|---|---|---|
| Input / 1M tokens | $10.00 | $9.00 | 10% OFF |
| Output / 1M tokens | $50.00 | $45.00 | 10% OFF |
| Cached read / 1M tokens | $0.25 | $0.225 | 10% OFF |
| Cached write / 1M tokens | $15.25 | $13.72 | 10% OFF |
| Context window | 1M tokens | 1M tokens | flat-rate |
Two honesty notes on this table before you use it. First, the “official list price” column is Modellix’s own label, and the input, output, and cached-read figures in it match Anthropic’s wording verbatim — “$10 per million input tokens and $50 per million output tokens” and cache reads “75% less, or $0.25 per million tokens” — but the cached-write figure of $15.25 is not stated on Anthropic’s announcement page, which never quotes a cache-write rate. What we can say is that Modellix’s price for every dimension sits 10% below the listed official price; that cache-write column is Modellix-listed, not an Anthropic-confirmed figure. Second, the cached-write price is shown here at the page’s rendered $13.72 (the underlying payload value is $13.725; we follow the page-render convention). And third, the discount is a same-day snapshot: on September 3 the same table showed Fable 5.1 at list price with no discount, and within 24 hours every Anthropic row moved to 10% OFF. Re-check the price page before you budget — the same way you’d re-check Anthropic’s — because neither the discount nor availability is promised to persist.
Fable 5.1 launched on September 1, 2026, per Anthropic’s Claude Fable page, and its rate card is otherwise identical to Fable 5’s — the $10/$50 headline did not change between generations. The change sits in the cache columns.
The four numbers on a Fable bill — and what each one means
A single successful API request can touch four different rates, and quoting “input/output” misses the two that decide whether long agent runs are affordable:
- Input — the tokens you send that are billed fresh. On Fable 5.1 that is $10 per 1M (Modellix: $9).
- Output — the tokens the model generates, billed at $50 per 1M (Modellix: $45). At five times the input rate, output is where verbose runs get expensive.
- Cached read — tokens the model reads from context it has already processed instead of reprocessing from scratch. Fable 5.1 charges $0.25 per 1M for those reads (Modellix: $0.225), replacing the base input rate for the reused portion.
- Cached write — the one-time cost of storing that context the first time it is processed, listed at $15.25 per 1M on Modellix’s table (Modellix: $13.72). It is the premium that makes every later read cheap.
We are describing what the four dimensions are, not how to make a provider cache anything — caching behavior is Anthropic’s to decide on its API, and the gateway bills what the model reports. Two things worth knowing structurally: Fable 5.1 is flat-rate, so the 1M context window does not raise the per-token price the way tiered models do (on tiered models like several OpenAI and xAI listings, input doubles past a 200K–272K threshold; Fable does none of that). And only successful responses are billed — usage-based billing means a failed request is not a surprise line item.
Illustrative Modellix artwork, not a product screenshot: the four billing streams a Fable bill is assembled from. Stream weights are conceptual; no numbers are shown.
Why Fable 5.1 bills less than Fable 5: the cache-read cut, in arithmetic
Anthropic’s own words, from the Fable 5.1 announcement: “Fable 5.1 will cost an estimated 25% less than Fable 5 for typical workloads, wherever usage is billed by token… For highly agentic work, the savings will often be much larger—up to approximately 45%.” The mechanism is a single line item: “Cache reads now cost 75% less, or $0.25 per million tokens.”
That “75% less” wording is the whole story — input and output prices did not move between Fable 5 and 5.1, so the read price is the only number that changed (75% less than the previous rate puts Fable 5’s cache reads at $1.00 per 1M, derived from Anthropic’s own arithmetic). Whether the savings reach 25% or 45% depends entirely on how read-heavy your workload is. Here is that relationship on a concrete, illustrative shape: a long agent run that reuses a ~100K-token prefix every turn, sends about 1M fresh input tokens and generates about 1M output tokens over the run (token buckets are illustrative, at official list rates):
| Run length (turns × 100K prefix) | Cache reads | Fable 5 (reads $1.00) | Fable 5.1 (reads $0.25) | Difference |
|---|---|---|---|---|
| 100 turns | 10M | $70.00 | $62.50 | −10.7% |
| 300 turns | 30M | $90.00 | $67.50 | −25.0% |
| 800 turns | 80M | $140.00 | $80.00 | −42.9% |
The 300-turn row lands exactly on Anthropic’s “around 25%” claim for typical workloads, and the 800-turn row approaches the “up to approximately 45%” figure for highly agentic, context-heavy work — because when cache reads dominate the bill, cutting their price 75% dominates the total. That is the arithmetic nobody on the current SERP has done: the guides ranking for this keyword still treat $10/$50 as if the cache columns did not exist. Whether Fable 5.1 also outperforms Fable 5 enough to matter is a capability question for Anthropic’s benchmarks, not this pricing page.
What it costs on Modellix: 10% OFF, and the price-page calculator that reproduces it
Put the 300-turn example through Modellix and the read-price story gets a second layer: the same tokens, billed at the actual Modellix rates ($9 input, $45 output, $0.225 cache read), come to $60.75 — exactly 10% below the $67.50 official-list total, because every dimension is discounted by the same 10%. The Discount column on the Modellix LLM price page prints “10% OFF” verbatim for anthropic/claude-fable-5.1, which makes this the rare row where a gateway’s price really is lower than the vendor’s sticker on every billing dimension. Two caveats, restated for emphasis: the 25%/45% savings above are Anthropic’s claims about its own Fable 5 → 5.1 pricing and are not a Modellix discount, and the 10% OFF is a snapshot taken September 4, 2026 — discounts on this table have already changed once within 24 hours this week, so treat it as current-until-rechecked.
Two things on that price page make the claim verifiable rather than rhetorical. First, the page ships a cost calculator: you enter input, output, cached-read, and cached-write token counts, and it prints one line per dimension plus two summary lines — the demo shown on the page today reads “Estimated Modellix cost $0.324 / Original price: approximately $0.36, approximately 10% less.” Enter the 300-turn example above (1,000,000 input, 1,000,000 output, 30,000,000 cached read, 0 cached write) and the calculator reproduces the same arithmetic from the displayed rates: Input $9.00, Output $45.00, Cached Read $6.75, Estimated Modellix cost $60.75, Original price approximately $67.50 — approximately 10% less. The calculator prices from the same displayed rates, so you can verify the totals yourself.
Second, the discount gains context when you see where Fable 5.1 sits in the catalog. It is the most expensive model on the 28-model table by output rate ($45 per 1M effective), and a few rows below it the same key reaches DeepSeek’s V4 Flash at $0.396 per million input tokens — the most expensive and one of the least expensive text models in the catalog share one API key, one billing pool, and one protocol surface, so “trying a cheaper model” is a one-string change (model: anthropic/claude-fable-5.1 → model: deepseek/deepseek-v4-flash), not a new account. For the full 28-model picture at one set of per-million rates, our cross-vendor LLM API pricing comparison puts every row side by side.
Verify the bill after the run: GET /v1/logs
A rate card tells you what a request should cost; the step every other Fable pricing page skips is confirming what it did cost — including whether the cache hits you are counting actually happened. On the Modellix LLM gateway, every request is logged and the LLM API guide documents the read endpoint: GET /v1/logs, which returns per-request records with a cost field and three token buckets — prompt_tokens, completion_tokens, and cached_tokens — plus the model as { provider, model_name } and an optional X-Mdlx-User-Id filter so you can attribute spend to individual end users (the header is optional, 8–128 characters, and the log endpoint filters on it). The read is a plain HTTP call, documented on the same page:
1 | curl -sS "https://llm.modellix.ai/v1/logs?start_time=1725300000&end_time=1725386400&mdlx_user_id=alice-prod-01" \ |
Why the cached_tokens bucket matters for this specific model: with Fable 5.1, the difference between a cache read and a fresh input read is $0.225 vs $9.00 per million tokens — 40×. If a long agent run was supposed to be saving via cached reads, the log tells you how many tokens actually billed at the read rate on each request, so the “25% cheaper” story stops being an estimate and becomes a sum you can audit after the run. Combined with the calculator on the price page, you can predict the bill before a run and reconcile it after one — the two halves of not being surprised by a Fable invoice.
Illustrative Modellix artwork: the model you are pricing sits on the same rail — and the same key — as cheaper models, and every request is logged against cost and token buckets. No numbers shown.
(The log covers the LLM text gateway; media generation on api.modellix.ai keeps its own logging endpoint, and the gateway’s host and protocol boundaries are documented in the Modellix LLM overview.)
Fable 5.1 on the Modellix LLM Gateway
Read the LLM API reference for model IDs, the /v1/logs request-log fields, and billing notes in the Modellix documentation.
View DocsFrequently Asked Questions
How much does Claude Fable cost?
The short answer to Claude Fable pricing in September 2026: Fable 5.1 costs $10 per million input tokens and $50 per million output tokens on Anthropic’s rate card, with cache reads at $0.25 per million tokens. On Modellix, the same Claude Fable 5.1 pricing lands at $9 input / $45 output / $0.225 cached read, and the Discount column reads “10% OFF”. Both Anthropic’s Claude Fable pricing and the Modellix table were read on September 4, 2026, and Fable 5.1 keeps Fable 5’s $10/$50 headline — the cache-read price is what dropped.
What context window does Claude Fable have?
The Claude Fable context window is 1M tokens on Fable 5.1, and the model is flat-rate — the window does not change the per-token price the way tiered models do. “Context in the table is the supported window,” as the pricing FAQ puts it; Fable 5.1’s Input Context column shows “—“ because there are no pricing tiers.
Can you use Claude Fable 5 for free?
No — and there is no free route to a Fable-class model. Claude Fable API pricing is strictly per-token: every successful response is billed from token usage, and there is no free tier for Fable-class models. New Anthropic developer accounts receive a small amount of free credits for testing, per Anthropic’s pricing documentation, but a finite trial credit is not a free API. On Modellix’s side, the $1 signup credit ended August 19, 2026 (trial credit is now by email request), and the two $0-priced models on the LLM page — modellix-ai/free-llm and zai/glm-4.7-flash — have nothing to do with Anthropic and carry no usage documentation; they are not a free way to call Claude.
Is Claude Fable 5.1 more expensive than Opus?
On the sticker, yes: Fable 5.1’s output rate is $50 per 1M (Modellix: $45) against Opus 5’s $25 per 1M (Modellix: $22.50, also 10% OFF) — roughly double on output, derived from the two rate cards on September 4, 2026. Whether double the output price is the cheaper run once cache reads, token efficiency, and task shape are included is exactly the question the Fable 5.1 vs Opus 5 decision guide answers; this page only gives you the numbers.
Is Claude Fable still available, or did 5.1 replace it?
Fable 5.1 is generally available, and it is an upgrade to the Fable line rather than a retirement of Fable 5 — Anthropic’s help center article on Fable models describes both as included in paid Claude plans at no extra cost on the subscription side. Its sibling Claude Mythos 5.1 is the same underlying model with different safeguards, available only through Anthropic’s trusted-access programs — it is not in Modellix’s catalog, and we won’t imply you can route to it. The subscription plan and the API are separate meters: plans bundle usage on Anthropic’s chat and coding surfaces, while API calls bill per token with no cap.
Will Fable 5.1 really cost less than Fable 5?
For token-billed usage, the mechanism is real and Anthropic’s: cache reads fell 75% to $0.25 per million tokens, and since input and output prices are unchanged, that read cut is the entire delta. Anthropic estimates typical workloads come in about 25% lower and highly agentic work up to about 45% lower. Those figures are Anthropic’s claims about its own billing — we reproduce the arithmetic above so you can see which workload shapes land where, but we don’t restate them as a Modellix discount.
Check Current Fable 5.1 Pricing on Modellix
Log in to see the live listed rate for anthropic/claude-fable-5.1, run the price-page calculator, and pull per-request cost logs on one key.
LoginPricing and model availability change without notice. Figures above were captured September 4, 2026 from the Modellix LLM price page, from Anthropic’s Fable 5.1 announcement, and from Anthropic’s model pricing documentation; the cache-write rate shown is Modellix’s listed figure, which Anthropic’s announcement page does not quote. Modellix is an API aggregator with a commercial interest in this guide and is not affiliated with Anthropic; the “10% OFF” is a same-day snapshot, so verify current rates against the live sources before committing spend. Access language models from Anthropic, OpenAI, Google, DeepSeek, Qwen, and more through a single API key at modellix.ai.
