Claude Fable 5.1 is Anthropic's updated frontier model, released September 1, 2026, and it's already live in Claude.ai, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry under the model ID claude-fable-5-1. Base token pricing hasn't moved from Fable 5, but cache reads are 75% cheaper, and Anthropic says that alone cuts costs up to 45% on heavily agentic workloads.
Short answer: Claude Fable 5.1 launched September 1, 2026, at the same $10/M input and $50/M output pricing as Fable 5, but cache reads dropped 75% to $0.25/M tokens. It keeps the 1M-token context window and adds a defensive vulnerability-discovery capability. Access it at claude.ai, through the API as
claude-fable-5-1, or via Bedrock, Vertex AI, and Microsoft Foundry.
In my testing, I pulled the pricing and benchmark figures below straight from Anthropic’s launch post and cross-checked the context window and spec details against the current models overview docs, since model launches tend to get restated sloppily by aggregator sites in the first day or two. Here's what Fable 5.1 actually costs, how it stacks up against Anthropic's other current models, and the exact steps to start using it.
What you'll need
For casual use, a free claude.ai account is enough — Fable 5.1 is available from the model picker the moment you sign in, no separate approval step. It replaces Fable 5 as the top option in that dropdown, though Anthropic is keeping Fable 5 available for now under its own redeployment commitment. For the API path, you need an account at console.anthropic.com, a payment method on file, and a token budget in mind — at $10/M input tokens, sending a 10,000-word document (roughly 13,000 tokens) once costs about 13 cents before output. If you're building anything with repeated context, plan around the cache pricing specifically, since that's where this release actually saves money.
Step-by-step: Claude Fable 5.1
1. Check which surface you're using it from
Claude.ai for chat, the Claude API for anything you're building, or Bedrock, Vertex AI, and Microsoft Foundry if your stack already runs through one of those clouds. Fable 5.1 is available on all of them at launch, so which one you use comes down to your existing setup, not availability.
2. Confirm the model picker actually shows 5.1, not 5
Open the model dropdown in claude.ai and look for "Fable 5.1" specifically. Anthropic restored access to the original Fable 5 after briefly pulling it earlier this year, so both versions are selectable right now, and it's easy to stay on the older one out of habit.
3. Pick your plan based on how often you'll lean on it
Fable 5.1 is priced identically to Fable 5 — $10 per million input tokens, $50 per million output tokens — so plan math from before still applies. Free lets you try it, but usage caps hit fast at this price point. Pro and Max both unlock heavier use; Max is the one to pick if Fable-class reasoning is your daily driver rather than an occasional escalation from a cheaper model.
4. Get an API key if you're building anything
Sign in at console.anthropic.com, generate a key, and call the model as claude-fable-5-1. Its reliable knowledge cutoff is June 2026, so attach a web search tool for anything more recent — same rule as every current Claude model.
5. Leave adaptive thinking on and set effort instead
Fable 5.1 always has thinking enabled — you can't pass thinking: {"type": "disabled"} the way you could on older Claude models. To control how much it reasons before answering, use the effort parameter (low through max), which defaults to high on both the API and Claude Code. In my testing, dropping to low or medium on a routine task produced results close to Fable 5's old high-effort output, which matches Anthropic's own claim that 5.1 does more per effort tier than its predecessor.
6. Watch cache-read costs on repeated-context workloads
If your integration re-sends the same system prompt, long document, or codebase context across calls, prompt caching is where Fable 5.1 actually gets cheaper than Fable 5 — cache reads fell from $1.00 to $0.25 per million tokens. Anthropic's own estimate is up to 45% lower total cost on highly agentic, cache-heavy work, versus roughly 25% on typical usage.
Example prompts you can copy
These lean on what Fable 5.1 is specifically built for — long-horizon reasoning and demanding agentic work — rather than generic chat requests:
- "Here's my full codebase context [paste or attach]. Trace this bug to its root cause instead of patching the symptom, and explain the actual mechanism." (Tests the "avoids shortcuts" improvement Anthropic highlighted at launch.)
- "Review this multi-step data pipeline for silent failure points, not syntax errors. Rank them by how likely each is to produce wrong output that looks correct." (Good use of the higher effort tiers.)
- "Here's a PDF with embedded charts and tables [attach]. Pull out the three numbers that matter most for this decision and flag anything the chart visually implies but the data doesn't support."
- "I'm running a multi-day agentic task against [tool/API]. Break it into checkpoints and tell me what you'd verify before trusting the output at each one."
- "Compare these two technical proposals section by section and tell me what's missing from each, not just what's present." (Long-context knowledge work Anthropic built Fable-class models for.)
Common mistakes to avoid
The first mistake I'd flag: assuming the model picker's "Fable" label means 5.1. Anthropic is keeping the original Fable 5 selectable alongside it, so check the version number, not just the family name. Second, running everything at default high effort when a routine task doesn't need it — you pay the same per-token rate regardless of effort level, so a higher tier mostly costs you time. Third, expecting Fable 5.1 to know about anything after its June 2026 cutoff without a search tool attached; it reasons confidently from stale information instead of flagging the gap. Fourth, ignoring prompt caching entirely — if your workload re-sends the same long context repeatedly, skipping caching means missing the actual price cut this release delivers. Fifth, defaulting to Fable 5.1 for tasks Opus 5 or Sonnet 5 handle just as well at a fraction of the cost; per Anthropic's own guidance, Fable-class models are for demanding reasoning and long-horizon agentic work specifically, not every request.
Claude Fable 5.1 vs. Anthropic's other current models
| Model | Input $/M tokens | Output $/M tokens | Cache read $/M tokens | Context window | Notable |
|---|---|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $50.00 | $0.25 | 1M tokens | Most capable current model; cache reads 75% cheaper than Fable 5 |
| Claude Opus 5 | $5.00 | $25.00 | $0.50 | 1M tokens | Best default for most agentic coding and enterprise work |
| Claude Sonnet 5 | $2.00 | $10.00 | $0.20 | 1M tokens | Best combination of speed and intelligence |
| Claude Haiku 4.5 | $1.00 | $5.00 | $0.10 | 200K tokens | Fastest current model |
Pricing confirmed against Anthropic’s launch post and current models docs on 2026-09-02. On Anthropic's own benchmark set, Fable 5.1 posts real gains over Fable 5 — Terminal-Bench-Science 0.1 more than doubled (52.6% vs. 24.7%), Terminal-Bench 4.0 coding rose to 55.8% from 42.0%, and OSWorld 2.0 (strict) climbed to 41.7% from 36.1%. Anthropic also reports it outperforming both Opus 5 and OpenAI's GPT-5.6 Sol across multiple benchmarks, though Opus 5 remains the cheaper default for work that doesn't need Fable-class reasoning.
Tools that make this easier
If you're deciding whether to switch your default model or just try Fable 5.1 alongside what you already use, my how to use Claude AI guide covers the account and plan setup this article assumes. I covered Anthropic's cheaper flagship in Claude Opus 5: pricing, benchmarks, and setup guide, useful if you want to know exactly when Fable-class pricing is and isn't worth it. If you're weighing Claude against the other major assistants, how to use ChatGPT and how to use Gemini cover the equivalent setup on each. For coding specifically, see best AI tool for code, and for a head-to-head on physical-world reasoning tasks, I ran GPT-5.6 Sol vs. Claude Fable 5 for physical AI before this update landed. My starter kit for AI is the place to start if you haven't set up any model accounts yet, and how we test AI tools explains how I verify numbers like these before publishing them.
Where this leaves you
Claude Fable 5.1 isn't a repricing move — the headline input and output rates are identical to Fable 5 — but the 75% cheaper cache reads make a real difference if your workload is agentic and context-heavy, and the benchmark jumps on coding and computer-use tasks are large enough to matter. If you're already paying for Fable-class reasoning, upgrading costs nothing extra and gets you a stronger model. If you've been avoiding Fable 5 on price, the math hasn't changed for a one-off prompt, but it has for anything that re-sends context repeatedly. Try it from the model picker on your next long-context or multi-step task before assuming your current default is good enough.
Last updated: September 2, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Frequently Asked Questions
Claude Fable 5.1: is it free?
You can try it on Claude's free plan, but usage caps hit quickly at this model's price point since Free has the smallest overall allowance. The API isn't free either way — it's pay-per-token at $10/M input and $50/M output, billed to a card on file at console.anthropic.com.
How long does it take to start using Claude Fable 5.1?
Through claude.ai, a couple of minutes — open the model picker and select "Fable 5.1." Through the API, expect 10–15 minutes to create an account, generate a key, and send a first test call with claude-fable-5-1 as the model string.
What is the easiest way to try Claude Fable 5.1?
Open claude.ai, select Fable 5.1 from the model picker, and run a real long-context or multi-step task through it rather than a short generic prompt — that's where the model's strengths and its price actually show up.
Is Claude Fable 5.1 better than Claude Opus 5?
On Anthropic's own benchmarks, yes, on the hardest reasoning and agentic-coding tasks — but Opus 5 costs half as much per token and is Anthropic's recommended default for most agentic coding and enterprise work. Fable 5.1 is worth the premium specifically for demanding, long-horizon reasoning tasks, not routine requests.
What's the difference between Claude Fable 5.1 and Claude Mythos 5.1?
They're the same underlying model. Fable 5.1 ships with Anthropic's standard safety classifiers and is available to everyone; Mythos 5.1 has fewer of those safeguards and is limited to U.S. organizations approved through Anthropic's Cyber Verification Program or Life Sciences Verification Program.