On June 9, 2026, Anthropic released two models with the same brain and very different front doors. Claude Fable 5 is the public one. Claude Mythos 5 is the restricted one. And if you pay for Claude, there's a deadline you should know about: Fable 5 is included in paid plans only until June 22, 2026.
We've been running our daily engineering on Fable 5 since launch day, so this guide covers what's verified, who gets what, what it costs, and what's actually worth doing with it before the window closes.
One Model, Two Names
Fable 5 and Mythos 5 are the same underlying model. The difference is safety classifiers.
Claude Fable 5 (API ID claude-fable-5) is what Anthropic calls a Mythos-class model "made safe for general use." It ships with classifiers that decline certain requests — offensive cybersecurity, biology and chemistry weapons research, and attempts to distill the model. When a classifier triggers, the query is answered by Claude Opus 4.8 instead, or refused outright. Anthropic says these safeguards fire in less than 5% of sessions on average.
Claude Mythos 5 (API ID claude-mythos-5) is the same model without those classifiers in some areas. It is not generally available. Access goes through Project Glasswing, Anthropic's collaboration with the US government, and is limited to vetted cyberdefenders, infrastructure providers, and selected biomedical researchers.
Both models share the same specs: a 1M token context window by default, up to 128K output tokens per request, and adaptive thinking that is always on.
Who Gets What
There are three paths, and they're very different:
- Paid Claude subscriptions (Pro, Max, Team, seat-based Enterprise): Fable 5 is included at no extra cost from June 9 through June 22, 2026. On June 23 it's removed from these plans.
- The Claude API and consumption-based Enterprise: Fable 5 is generally available right now, with no end date — on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry.
- Project Glasswing partners: Mythos 5, invite-only. If you think you qualify, the path is through your Anthropic, AWS, or Google Cloud account team.
Free-tier users get neither. That's a real shift from previous launches, where new flagship models eventually trickled down to free accounts.
The June 22 Catch
This is the part most coverage buries. If you're on a paid plan, Fable 5 works in your normal Claude apps today at no extra cost — but only until June 22. After that, continued use requires usage credits billed at standard API rates once your plan limits are exhausted.
Anthropic's own framing is telling: "when sufficient capacity allows us to do so, we aim to restore Fable 5 as a standard part of subscription plans." Read between the lines and the two-week window looks like a free trial that doubles as a load test. They want to know what demand looks like before committing the capacity permanently.
Practical takeaway: treat June 9–22 as your evaluation window. If Fable 5 meaningfully changes your workflow, you'll know by the 22nd, and you can decide whether API pricing is worth it for your use case.
What It Costs
On the API, Fable 5 and Mythos 5 are priced identically: $10 per million input tokens and $50 per million output tokens. Anthropic frames this as less than half the price of the earlier Claude Mythos Preview; press coverage frames it as double the price of Claude Opus 4.8. Both framings are accurate — it depends on which direction you're coming from.
For context on choosing between Claude models by cost and capability, our Sonnet vs Opus guide covers the decision framework, and it still applies: most production workloads don't need the biggest model. Fable 5 is for the work where the biggest model is the whole point.
What Changes in the API
If you integrate Fable 5, four behaviors are new compared to Opus, Sonnet, and Haiku:
- Refusals are HTTP 200, not errors. When a classifier declines a request, the Messages API returns
stop_reason: "refusal"as a successful response, and reports which classifier fired. You are not billed for a request refused before any output is generated. - Server-side fallback exists. A
fallbacksparameter (in beta) lets the API retry a refused request on another Claude model for you. There's also SDK middleware for client-side retries, and fallback credit refunds the prompt-cache cost of switching models so you don't pay it twice. - Thinking is always on. Adaptive thinking is the only mode —
thinking: {"type": "disabled"}is not supported. You control depth with the effort parameter instead. This pairs with the advice in our temperature settings guide: the new generation of Claude models gives you fewer raw dials and more intent-level controls. - Raw chain of thought is never returned. Thinking blocks contain either a readable summary (
"summarized") or nothing ("omitted", the default).
One more operational note: both models carry 30-day data retention and are not available under zero data retention. If your compliance posture requires ZDR, Fable 5 is off the table for now.
Is This Fair?
Honest answer: mostly, with two real critiques.
The tiered structure itself — everyone gets nothing for free, paying users get a window, vetted organizations get the unrestricted version — is defensible. Anthropic published the safeguard behavior, the fallback rates, and the access criteria instead of quietly degrading the public model. A model that declines under 5% of sessions and tells you exactly which classifier fired is a more honest contract than most of the industry offers.
Critique one: the pricing pushes out independent developers. At $10/$50 per million tokens, an agentic coding session that chews through a large codebase gets expensive fast. The two-week subscription window gives indie builders a taste of capability they may not be able to afford on June 23. If Anthropic restores subscription access, this objection mostly dissolves; if it doesn't, the most capable widely released model becomes an enterprise product in practice.
Critique two: Mythos 5 is an accountability question. Government-adjacent organizations get a frontier model with safeguards lifted, and the public gets no way to audit how it's used. There are good reasons for cyberdefenders to have the unguarded version — attackers won't be using the guarded one. But "trust the vetting process" is doing a lot of work in that sentence, and it's fair to want more transparency about who is in Project Glasswing and what they're allowed to do.
The Best Things to Do With Fable 5 Right Now
If you have paid-plan access until June 22, don't waste the window on chat. Fable 5's edge over Opus is long-horizon work. The highest-value uses we've found:
- The migration you've been putting off. Anthropic's launch case study is a complex Ruby codebase migration completed in a day that would otherwise have taken a team over two months. That matches our experience: Fable 5 holds a plan across hours of agentic work without losing the thread. Pick your scariest refactor and run it now.
- Whole-codebase analysis in one context. The 1M token window fits most real repositories in a single prompt. Architecture reviews, dead-code audits, security passes, dependency-upgrade planning — the quality jump from "file-by-file" to "everything at once" is significant.
- Deep research with effort control. Adaptive thinking plus the effort parameter means you can ask for genuinely hard reasoning — literature synthesis, competitive analysis, multi-document contract review — and pay the thinking cost only when the question deserves it.
- Long agentic sessions in Claude Code. This is where we live. Fable 5 in agentic loops handles multi-step builds — research, write, test, deploy, verify — with noticeably less drift than previous models. If you've set up custom instructions, they carry further because the model holds them across much longer sessions.
How to Get Mythos 5 Access
Short version: you probably don't, and that's by design. Mythos 5 is limited to approved Project Glasswing customers — cyberdefenders and infrastructure providers, plus selected biomedical researchers with the biology safeguards removed. Anthropic has said a broader trusted-access program is planned, with expansion to biology researchers first and eventually a wider cybersecurity program.
If your organization does defensive security or biomedical research at serious scale, contact your Anthropic, AWS, or Google Cloud account team. Everyone else: Fable 5 has the same capabilities for everything outside the safeguarded domains, and it's available today.
Bottom Line
Fable 5 is the most capable model Anthropic has ever made generally available, and for thirteen days it's sitting in paid Claude plans at no extra cost. Use the window deliberately: run the big migration, do the whole-repo audit, stress-test it on your hardest real work. By June 22 you'll know whether it's worth API pricing to you — and you'll have gotten your hardest backlog item done either way.