Claude Mythos vs Claude Fable 5: What’s the Difference?

Anthropic launched the Claude 5 generation with two names on one model, and the naming confuses almost everyone. The short version: Claude Fable 5 and Claude Mythos 5 are the same underlying model — the first release of Anthropic’s new Mythos class, a capability tier above Opus. Fable 5 is the one you can actually use: generally available, with additional safeguards around dual-use capabilities. Mythos 5 is, in Anthropic’s own words, “the same underlying model as Fable 5, but with the safeguards lifted in some areas” — and it is only available to approved organizations.

What is Claude Fable 5?

Fable 5 is the first model in the Claude 5 family and the debut of the Mythos class, which Anthropic positions above its Opus class in capability. The announcement’s exact claim: “Fable 5’s capabilities exceed those of any model we’ve ever made generally available.” It launched everywhere at once — the Claude API plus the Pro, Max, Team and Enterprise plans — at $10 per million input tokens and $50 per million output tokens. That output price makes it the most expensive Claude by far: double Opus 4.8, more than triple Sonnet.

The “generally available” part comes with a condition. Because the model is strong in dual-use domains, Fable 5 ships with additional safety measures: classifiers watch for misuse in cybersecurity, biology and chemistry, and for attempts to distill the model, and per Anthropic, flagged requests fall back to Claude Opus 4.8 instead. For ordinary coding and writing work you are unlikely to meet these classifiers at all — but they are the entire difference between the two names.

What is Claude Mythos 5?

Claude Mythos 5 is that same model with some of those safeguards removed. Anthropic describes it as the same underlying model as Fable 5 with “the safeguards lifted in some areas”, initially deployed through Project Glasswing — a program run in collaboration with the US government for cybersecurity professionals — with plans to expand through a broader trusted-access program. The announcement also mentions a separate biology-focused program offering Fable 5 with the biology and chemistry safeguards removed. In other words: Mythos 5 is not a bigger or smarter model. It is the same intelligence with fewer restrictions, for vetted organizations whose work needs them lifted.

Can you get access to Claude Mythos?

• Claude Fable 5 — yes, today. Any API account or paid Claude plan can use it; it is the default answer for anyone asking to use “the Mythos-class model”.
• Claude Mythos 5 — not publicly. Access runs through Anthropic’s approved-organization programs (Project Glasswing today, a broader trusted-access program announced as planned). There is no public signup or self-serve tier described in the announcement.

Claude Mythos benchmarks: what can actually be verified

Because Mythos 5 is restricted, there are effectively no independent public Mythos benchmarks — treat any you see with suspicion. But since Anthropic states it is the same underlying model, Fable 5’s public results are the closest observable proxy for everyday work like coding. And Fable 5 we can measure: it runs in our arena on the same one-shot prompts as every other model, next to Opus 4.8, Sonnet, GPT-5.5, Gemini and the open-weight field.

The receipts so far: Fable 5 runs at five thinking-effort levels in our roster, and as of July 7, 2026 it holds the second-most community votes overall — behind Opus 4.8 at high effort, ahead of everything else. The live leaderboard has the current tally, cost-per-task and generation times; the full Fable 5 report digs into where its token appetite pays off and where it does not.

Try it yourself: Fable 5 vs Opus, live

You cannot A/B Mythos 5 against anything. You absolutely can watch Fable 5 and Opus 4.8 build the same app from the same prompt — this site started life as exactly that head-to-head and grew into a 20-family arena. Open the coding arena, put Fable 5 next to Opus in compare mode, judge blind, and vote. If you are choosing a day-to-day coding model rather than chasing the flagship, our full breakdown of the best AI for coding covers the whole field, including what $50 per million tokens does and does not buy you.