Claude Fable 5 Explained: Pricing and Capabilities
Claude Fable 5 is Anthropic's most powerful public model yet. See its benchmarks, $10/$50 pricing and how its API cost compares to GPT, Gemini and Grok.
On this page
Update — 12 June 2026: Anthropic has suspended Claude Fable 5 and Mythos 5 for all users after a US government export control directive. Every other Claude model still works. For what happened and the best alternatives to switch to now, see Claude Fable 5 and Mythos 5 suspended.
Claude Fable 5 is Anthropic's most powerful public model, released on 9 June 2026. It shares its weights with the restricted Mythos 5 but adds safeguards. It costs $10 per million input tokens and $50 per million output tokens, which puts it at the top end of the market.
Anthropic does not normally give the public its frontier "Mythos" models, but Fable 5 is the exception. For the first time, anyone can call this tier of power through a normal API, and the benchmark scores have made it the biggest launch of the year so far.
But power is only half the story. At $10 in and $50 out per million tokens, Fable 5 is also one of the priciest flagship models you can buy, which makes cost the question you cannot skip.
This guide covers what Fable 5 is, what it does well, what it costs, and how that price compares to GPT, Gemini and Grok. To model your own spend now, jump straight into the live cost calculator.
What Is Claude Fable 5?
Claude Fable 5 is a large language model from Anthropic that takes text and images as input and returns text. Anthropic calls it state of the art on nearly every benchmark it tested, with its biggest strengths in coding, knowledge work, science and vision.
In plain terms, it is built for the hardest reasoning and coding tasks we ask of AI today.
What makes Fable 5 unusual is its lineage. Anthropic has warned that its top internal models are powerful enough to be risky in the wrong hands, so it keeps that "Mythos" tier locked down.
Fable 5 is the safe, public version of that same technology, and you can read the company's own framing in its Fable 5 and Mythos 5 announcement.
How Fable 5 Relates to Mythos 5
Fable 5 and Mythos 5 run on the same weights. The difference is access and safety, not brains:
- Mythos 5 is the restricted model, saved for vetted use under tight controls.
- Fable 5 is the public release of those weights, wrapped in safeguards that limit a few high risk topics.
This matters, because the model you can call today is truly frontier grade. You are not getting a watered down copy. You get the same brain, with a few doors locked. Both models cost the same, at $10 per million input tokens and $50 per million output tokens.
Claude Fable 5 Capabilities and Benchmarks
Anthropic frames Fable 5 as a leap, not a small step. The published scores back that up. Several are the highest ever reported for a public model.
Coding and Agentic Performance
Coding is where Fable 5 looks strongest. By Anthropic's figures, the model scores:
- 95.0% on SWE-bench Verified, a test of fixing real GitHub issues. You can read the method at SWE-bench.
- 80.3% on SWE-bench Pro, about 11 points ahead of the next best model.
- 88.0% on Terminal-Bench 2.1, which scores agentic, multi step command line tasks.
- 59.0% on Humanity's Last Exam, a very hard reasoning test.
An 11 point lead on SWE-bench Pro is huge. Frontier models normally sit within a point or two of each other. For teams building coding agents, that gap is the main reason to pay more.
State-of-the-Art Vision
Fable 5 is also the best vision model Anthropic has shipped. It reads images alongside text, and it can do far more than caption a picture.
Anthropic shows two examples. The model can pull exact numbers from detailed science charts. It can also rebuild a web app's source code from a screenshot. If your work involves charts, diagrams, documents or screenshots, this is a real feature, not a gimmick.
Claude Fable 5 Pricing Explained
Fable 5 uses simple per token pricing. No subscription is needed for API access:
- Input: $10 per million tokens.
- Output: $50 per million tokens.
Tokens are chunks of text, and as a rough guide three quarters of a word equals one token. Input is everything you send, such as your prompt, instructions and any documents, while output is everything the model writes back.
Because output costs five times more than input, jobs that write long answers cost far more than jobs that mostly read and sort.
There is also a launch window to plan around. Anthropic offers Fable 5 free to users on its Pro, Max, Team and Enterprise plans until 22 June 2026. After that, access moves to paid compute credits. If you are testing Fable 5, run your heaviest jobs before that date. The latest details sit in the Anthropic documentation.
Fable 5 API Cost Compared to GPT, Gemini and Grok
Pricing only means something in context. Here is how Fable 5 sits next to the other flagship models in 2026. All figures are per million tokens.
| Model | Input (per 1M) | Output (per 1M) |
|---|---|---|
| Claude Fable 5 | $10.00 | $50.00 |
| GPT-5.4 | $2.50 | $15.00 |
| GPT-5 | $0.625 | $5.00 |
| Gemini 3.1 Pro | $2.00 | $12.00 |
| Gemini 3 Flash | $0.50 | $3.00 |
| Grok 4.1 | $0.20 | $0.50 |
The pattern is clear. Fable 5 costs about four times more on input than GPT-5.4 and Gemini 3.1 Pro, and it costs far more than the budget tiers.
Grok 4.1 is the cheap outlier, at $0.20 input and $0.50 output. That makes Fable 5 about 50 times pricier on input and 100 times pricier on output. You pay for the lead in power, not for value per token.
For a full, up to date breakdown across every provider, see the providers comparison on CompareAI.
A Worked Cost Example
Headline rates feel abstract, so picture a steady workload of 10 million input tokens and 2 million output tokens a month. That suits a busy assistant or a small set of agents. The monthly cost works out like this:
| Model | Monthly cost (10M in, 2M out) |
|---|---|
| Claude Fable 5 | $200 |
| GPT-5.4 | $55 |
| Gemini 3.1 Pro | $44 |
| Grok 4.1 | $3 |
The same job costs about $200 a month on Fable 5. On the mid tier rivals it costs roughly $44 to $55. The choice is rarely about whether you can afford $200.
It is about value. Does Fable 5's edge produce enough extra, in shipped code or correct answers, to justify paying three to four times more? Model your own numbers in the cost calculator before you commit.
The Safeguard Fallback and What It Means for Your Costs
Fable 5 carries Mythos level power, so Anthropic guards a narrow band of high risk topics. Requests that touch cybersecurity, biology, chemistry or model distillation route to Claude Opus 4.8 instead. Anthropic has set these safeguards cautiously, and reports they trigger in fewer than 5% of sessions.
For most users this is invisible. But it has two practical effects worth knowing:
- Power: In guarded topics you get an Opus 4.8 answer, not a Fable 5 one. So do not expect frontier results on sensitive science or security prompts.
- Cost planning: The fallback swaps the model that handles the request, which can change the quality you pay for. If your app lives in those topics, test for the fallback, not just on open prompts.
This is a real gap in most launch coverage. It tends to quote the scores without noting that they apply to the unguarded paths.
Who Should Use Claude Fable 5?
Fable 5 is not the default pick for every project. It fits when power is the bottleneck and the value clearly beats the cost. Strong fits include:
- Complex coding: Coding agents and automated bug fixing, where the SWE-bench lead means more issues solved.
- Vision heavy work: Reading science charts, rebuilding interfaces, or processing document and chart images with high accuracy.
- High value knowledge work: Legal, financial or research analysis, where one correct answer is worth far more than its token cost.
For high volume, simple work, a cheaper model usually wins. Tasks like basic sorting, summaries or chat run almost as well on Gemini 3 Flash or Grok 4.1, for a fraction of the cost.
How to Access Claude Fable 5
There are two main routes to Fable 5 today:
- Claude plans: Pro, Max, Team and Enterprise users get Fable 5 free until 22 June 2026. After that, access moves to paid compute credits.
- API and platforms: Developers can call Fable 5 through the Anthropic API at the listed rates. It is also rolling out across major platforms and cloud marketplaces.
Either way, the economics come down to your token volumes. So estimate your spend before you build.
Claude Fable 5 Frequently Asked Questions
Is Claude Fable 5 the Same as Mythos 5?
Almost. Fable 5 and Mythos 5 run on the same underlying weights, so the raw intelligence is the same. The difference is that Fable 5 is the public release and adds safeguards, while Mythos 5 stays restricted to vetted users. Both cost $10 per million input tokens and $50 per million output tokens.
Why Is Fable 5 So Expensive Compared to Other Models?
You are paying for a clear lead in capability, not for value per token. Fable 5 tops the field on coding and vision benchmarks, and that edge carries a premium. For routine, high volume tasks, cheaper models like Grok 4.1 or Gemini 3 Flash deliver similar quality at a tiny fraction of the cost.
How Much Does Claude Fable 5 Cost per Month?
It depends entirely on your token volume. A steady workload of 10 million input tokens and 2 million output tokens costs about $200 a month. The same job runs at roughly $44 to $55 on Gemini 3.1 Pro or GPT-5.4. Use the cost calculator to model your own usage.
Fun Fact: Fable 5 Beat Pokemon From Pixels Alone
To show off its vision, Anthropic shared a timelapse of Fable 5 playing and beating Pokemon FireRed. The model used nothing but the on screen image. It read the game purely from pixels, the way you would by glancing at a screen. It is a neat reminder that Fable 5 sees images, not just text.
Final Verdict: Is Fable 5 Worth the Price?
Claude Fable 5 is the most powerful model the public has ever been able to call. On coding and vision, it sets a new bar. That power comes at a real premium.
At $10 input and $50 output per million tokens, it costs several times more than strong mid tier rivals. It costs many times more than budget options like Grok 4.1.
So is it worth it? It is worth the price when power is the constraint, and hard to justify when it is not. If you build complex agents, do serious vision work, or run tasks where a correct answer is worth far more than its token cost, the price is easy to defend.
If your workload is high volume and routine, your money goes further elsewhere. The smart move is to compare real numbers for your own use case, not the headline rate. You can do that on CompareAI, where pricing for all 10 leading providers stays up to date in one place.
