Mythos 5 and Fable 5: My Honest Take
By Miguel Torrez, founder of Genial Agency
You will know what Anthropic's Mythos 5 and Fable 5 change, what they cost, the compliance catch, and how to try Fable 5 yourself during the launch window. Then you can decide whether it fits your work or whether Opus 4.8 or Sonnet 4.6 is enough.
What you need
A Claude account (I find the $20 plan borderline impossible for Fable 5; plan on the $100 plan at least)
The latest Claude desktop app or Claude CLI
A demanding task to test it on, such as a cybersecurity or heavy back-end project
Step by step
Understand what changed
Mythos 5 and Fable 5 are a new model family that sits above Opus. Mythos was private for a few companies since Project Glasswing, and a Mythos 5 preview is still limited to a few companies. Fable 5 is the same class as Mythos with more restrictions, making it safer for general use.
Check the price before you switch
Output costs $50 per million tokens, about twice the price of Opus 4.8 and its predecessors. Expect it to consume large amounts of tokens. For most business work, Opus 4.8 and Sonnet already get the job done.
Read the benchmarks with care
Anthropic publishes these benchmarks about its own model, so take them with a grain of salt. Gains over the previous models are small across the board, except cybersecurity, where Fable 5 shows nearly a 2x jump over Opus 4.8. That is where I see its real value.
Weigh longer tasks against cost per task
The models can work independently for longer on a task, which mostly helps developers. The average cost per task also goes up, so factor that in before you hand it long jobs.
Note the vision and context gains
Vision improved, which helps front-end, design, image and video analysis work (the demo shows Fable 5 beating Pokemon Fire Red using computer vision). The context window did not grow, but the model is better at remembering what you told it within one conversation.
Decide if you are the audience
These models lean toward researchers, cybersecurity developers and heavy Claude power users. The main difference between the two: Fable 5 has more safeguards than Mythos 5. Both appear to have the same pricing and level of usage.
Check your compliance needs
Mythos class models require 30-day data retention for all traffic. If you need zero data retention, for example to comply with ISO 27001, you cannot use Fable 5 or Mythos 5 yet.
Test it while subscription access lasts
Every user gets Fable 5 access for now, but after a limited window (about 13 days at the time of the video) it moves to usage pricing only. If you want to test it or have a big cybersecurity project waiting, use your tokens during that window.
Switch to Fable 5 in Claude
Update the Claude desktop app and accept the prompt that tells you that you can switch to Fable 5. In the Claude CLI, update it and you will get the same prompt. It burns about twice the tokens of Opus 4.8, so watch your usage.
Claude (web and desktop download)https://claude.ai
Watch out for
Getting hooked on Fable 5 during the included window: it switches to usage pricing only and can burn a hole in your wallet.
Running it on a $20 plan: at twice the token burn of Opus 4.8, I call this borderline impossible.
Using it where you need zero data retention (such as ISO 27001): Mythos class models require 30-day retention.
Trusting the benchmark charts at face value: the company that made the model also made the benchmarks.
Watch next
6:59Claude Fable 5 Is Here to Stay - Maximize Your UsageLet Fable 5 plan while Opus subagents do the work so your allowance lasts longer.
5:44Use Fable 5 for 3D Websites Before It's GoneTest Fable 5 on a real build by prompting it to make a scrolling 3D website.
7:31How to Choose the Right Claude ModelDecide when Opus or Sonnet is enough for your everyday business work.
