The claude fable 5.1 cache read pricing is $0.25 per million tokens — a 75% cut from Claude Fable 5's $1.00 per million — while input stays at $10 and output at $50 per million tokens, according to Anthropic's official Claude Fable 5.1 documentation published with the model's release on 1 September 2026. That one line is the whole story of this model's economics: Anthropic did not discount the model, it discounted the part of the bill that dominates real agentic workloads. Cache reads on Fable 5.1 now cost 2.5% of the input price, versus the 10% that applies across the rest of the Claude lineup, and if your agents re-read long system prompts, tool definitions and conversation history on every turn — which is exactly what agents do — that difference compounds into the largest effective price cut Anthropic has shipped this year.
📺 Watch: NEW Claude Fable & Mythos 5.1 is ABSURD!
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside · Want AI SEO help 1-on-1? Book a free SEO strategy session →
Claude Fable 5.1 Cache Read Pricing: The Full Table
Here is the complete Fable 5.1 price list from the official model documentation, with the cache lines that matter highlighted against the context you need to judge them:
| Item | Claude Fable 5.1 | Claude Fable 5 |
|---|---|---|
| Input | $10 / MTok | $10 / MTok |
| Output | $50 / MTok | $50 / MTok |
| Cache read | $0.25 / MTok (2.5% of input) | $1.00 / MTok (10% of input) |
| 5-minute cache write | $12.50 / MTok | $12.50 / MTok |
| 1-hour cache write | $20 / MTok | $20 / MTok |
| Batch API | 50% off input and output | 50% off input and output |
| Context window | 1M tokens | 1M tokens |
| Max output | 128K tokens | 128K tokens |
Everything except the cache read line carried over unchanged from Fable 5. The model ID is claude-fable-5-1, it runs adaptive thinking always-on with a default effort of high, and Anthropic's docs commit to no retirement sooner than 1 September 2027 — relevant if you are budgeting a year ahead.
Why Cache Reads Dominate an Agent's Bill
Prompt caching lets the API store the stable front of your prompt — system instructions, tool definitions, reference documents, accumulated conversation — and re-serve it at the cache read rate instead of the full input rate on every subsequent call. An agent taking fifty tool-use turns re-sends that stable context fifty times. At Fable 5's $1.00 cache reads, a cached 200K-token context cost roughly $0.20 every turn, about $10 over those fifty turns. At Fable 5.1's $0.25, the same fifty turns cost about $2.50. Nothing about the workload changed; the meter did.
That is why the flat $10/$50 headline price is misleading in both directions. VentureBeat's launch report, quoting Anthropic's figures, puts the practical saving at roughly 25% for typical workloads and up to about 45% for highly agentic ones — the longer your sessions and the fatter your cached context, the closer you get to the top of that range. For the practical craft of keeping context lean in the first place, the guide to reducing Claude Code token usage pairs naturally with this price cut: trimming what you cache and paying 75% less to re-read it stack multiplicatively.
If you want your AI agent bills going down while output goes up — with the cost-optimised workflows already mapped — check out the AI Profit Boardroom → join and get the full Agent OS stack. Want your setup costed 1-on-1? Book a free SEO strategy session.
📺 Watch: Claude Fable 5.1 Just Set a New AI Performance Record
How Fable 5.1 Cache Economics Compare Across the Claude Lineup
The 2.5% cache read rate is unique to Fable 5.1 (and its invitation-only sibling Mythos 5.1); every other current Claude model reads cache at 10% of its input price. From the official pricing documentation:
| Model | Input / Output | Cache read | Cache read as % of input |
|---|---|---|---|
| Claude Fable 5.1 | $10 / $50 | $0.25 | 2.5% |
| Claude Opus 5 | $5 / $25 | $0.50 | 10% |
| Claude Sonnet 5 | $2 / $10 | $0.20 | 10% |
| Claude Haiku 4.5 | $1 / $5 | $0.10 | 10% |
Read that middle column twice, because it contains a genuine surprise: Fable 5.1 — the most expensive model in the lineup on paper — now reads cached context at half the price of Opus 5, and only 25% above Sonnet 5. On a long-running agent whose bill is mostly cached re-reads, the effective gap between "the frontier model" and "the sensible mid-tier model" collapses dramatically. Where the gap does not collapse is fresh input and, above all, output: Fable 5.1's $50 output rate is double Opus 5's $25 and five times Sonnet 5's $10, and no cache discount touches output. The Claude Opus 5 vs Fable 5 comparison covers the capability side of that trade-off, while Kimi K3 vs Fable 5 and GPT-5.6 vs Fable 5 put the pricing next to the strongest outside rivals.
What This Means for Your Costs in Practice
A few concrete implications follow directly from the numbers:
- Long-horizon agent sessions are the winning use case. The docs position Fable 5.1 for demanding reasoning and long-horizon agentic work, and the pricing agrees: the more turns re-reading the same cached context, the more the 2.5% rate pays you back.
- Cache writes did not get cheaper. At $12.50 per million for the 5-minute tier and $20 for the 1-hour tier, writing cache still costs more than plain input. Caching pays when contexts are re-read repeatedly — it costs you money on one-shot calls.
- Output remains the luxury item. At $50 per million, verbose responses are where budgets die. Cheap cache reads reward agents that think over a large stable context and answer tersely.
- Batch still stacks. The 50% batch discount on input and output applies as before, so non-urgent bulk work on Fable 5.1 gets both levers at once.
If your workloads run through Claude Code rather than the raw API, the same caching logic drives your costs there, and the Claude Code cost optimisation guide covers the operator-level tactics — while the free Claude Code guide maps the no-budget routes into the tooling. For choosing when Fable 5.1's premium is actually worth paying over a cheaper brain, the Goldie Bench write-up covers how these models compare in hands-on tests across real agent tasks.
A Worked Example: One Agent Day on Each Meter
Numbers make the claude fable 5.1 cache read pricing concrete. Take a single busy agent session: a 150K-token stable context (system prompt, tools, project files) written to cache once, then re-read across 40 turns, with 2K tokens of fresh input and 1K tokens of output per turn. On Fable 5, the cache reads alone were 150,000 tokens × 40 turns × $1.00 per million — about $6.00. On Fable 5.1 the identical re-reads cost about $1.50. The fresh input (80K tokens, $0.80), the one-hour cache write (150K tokens, $3.00) and the output (40K tokens, $2.00) are priced identically on both models, so the session drops from roughly $11.80 to $7.30 — a 38% saving from one line item, right in the band Anthropic's reported 25-45% estimate predicts for agentic work. Scale that to a team running dozens of such sessions daily and the cache line stops being a rounding error and becomes the budget decision.
The Verdict on Fable 5.1's Pricing
Claude Fable 5.1's cache read pricing is the quiet headline of the 1 September release. The sticker price says "same as Fable 5"; the meter says "materially cheaper for exactly the workloads this model is built for". Anthropic effectively repriced long-running agents without repricing the model — a distinction that rewards people who understand their own token mix. If your usage is chat-shaped, one-shot and output-heavy, nothing changed for you, and a cheaper model remains the right call. If your usage is agent-shaped — big stable contexts, many turns, structured outputs — Fable 5.1 just became significantly cheaper to run than its list price suggests, and the right move is to re-run your own cost maths before assuming the frontier tier is out of budget. That maths, not the benchmark chart, is where this release actually moves money — and it is exactly the kind of decision the Agent OS approach systematises: pick the brain per workload, cache aggressively, and measure what each layer of your stack really costs.
If you want frontier-model results without frontier-model bills — with the cost playbooks and daily tutorials to get there — check out the AI Profit Boardroom → get inside now. And if you want eyes on your exact setup, book a free SEO strategy session — bring your usage, leave with a plan.
