For months, Anthropic had an AI it refused to show anyone.
The model was called Mythos. It was too powerful to release publicly — capable of finding thousands of security vulnerabilities in major software, discovering zero-day exploits across operating systems and browsers, and doing things that made Anthropic's own safety team nervous. So they kept it locked away, sharing it only with 150 carefully vetted organizations under a classified program called Project Glasswing.
On June 9, 2026, that changed.
Anthropic launched Claude Fable 5 — a version of Mythos that's been made safe for general use. The AI they were afraid to release is now available to anyone with a Claude subscription. And the benchmarks it's posting are unlike anything we've seen from a publicly available model.
Here's everything you need to know — without the jargon.
What Is Claude Fable 5, Exactly?
Let's clear up the naming confusion first, because "Fable 5" is a genuinely weird name and most articles aren't explaining why it exists.
Claude Fable 5 is a Mythos-class model that Anthropic has made safe for general use. Think of it this way: Mythos is the engine, and Fable 5 is the road-legal version of that engine — same power underneath, but with safety systems that prevent it from doing dangerous things.
Fable 5 and Mythos 5 are the same model underneath. The practical difference is that Fable 5 ships with safety classifiers that watch every session and reroute high-risk queries to Claude Opus 4.8 instead of answering directly. Mythos 5 has those classifiers lifted in specific areas.
So when you ask Fable 5 something involving cybersecurity, biology, or chemistry at a dangerous level — it doesn't answer. It hands off to Opus 4.8 instead. According to Anthropic, the classifiers trigger in fewer than 5% of sessions on average. For most users, most of the time, you're getting the full Mythos experience.
The name "Fable" comes from the Latin fabula — "that which is told." Anthropic is telling a story with this release: powerful AI can be made safe for everyone, not just vetted organizations.
How It Fits Into Anthropic's Model Family
Before Fable 5, here's how Claude's model hierarchy looked:
Haiku → Fast and cheap. Good for simple tasks. Sonnet → The balanced middle. Most popular for everyday use. Opus → The previous flagship. Best publicly available model. Mythos → Secret. Restricted. Only for Project Glasswing.
Now there's a new tier:
Fable 5 → Mythos-class. Publicly available. The most powerful Claude ever released.
Mythos-class models sit above Opus in Anthropic's capability hierarchy. This isn't a minor update to Opus. It's an entirely new tier of capability — and it just became accessible to regular users for the first time.
The Benchmark Numbers That Stopped the AI World
Benchmarks can be boring. These ones aren't.
Claude Fable 5 scores 80.3% on SWE-Bench Pro, compared with 58.6% for GPT-5.5 and 54.2% for Gemini 3.1 Pro. This gives Fable 5 a 21.7-point lead over GPT-5.5 and a 26.1-point lead over Gemini on this Python-focused coding benchmark.
To understand why that gap matters: SWE-Bench Pro tests whether an AI can actually fix real GitHub issues from actively-maintained codebases. Not toy problems. Not textbook examples. Real bugs, real repositories, real complexity.
The gap between Fable 5 and Anthropic's own Opus 4.8 — more than 11 points — is larger than the gap between Opus 4.8 and Google's Gemini 3.1 Pro.
Let that sink in. The jump from the previous best Claude to Fable 5 is bigger than the entire gap between Opus 4.8 and Gemini 3.1 Pro. That's not incremental progress. That's a generational leap in one release.
The real-world proof: Stripe ran a 50-million-line Ruby codebase migration using Fable 5 in one day. The same task would have taken a full engineering team approximately two months. One day versus two months — that's not a percentage improvement. That's a different category of capability.
What Fable 5 Is Actually Great At
🧑💻 Coding — Miles Ahead of Everything Else
This is Fable 5's defining strength. Claude Fable 5 also posted the highest score among frontier models on Cognition's FrontierCode eval, a separate independent measure of frontier coding ability.
More importantly: The pattern in the results — biggest gains on the longest, hardest tasks — suggests the improvement is concentrated precisely where agentic workloads live.
This means Fable 5 doesn't just answer coding questions better. It can run autonomously across a large codebase for hours, making changes, fixing errors, and finishing tasks that would have required constant human supervision before. Claude Code now runs on Fable 5 — and the difference is immediately noticeable for anyone building software.
📚 Knowledge Work and Research
Fable 5 tops knowledge work with GDPval-AA score of 1932 versus 1769 for GPT-5.5 and 1314 for Gemini 3.1 Pro.
For professionals who use AI for research, writing, analysis, or document work — lawyers, consultants, researchers, writers — this matters. The model isn't just faster at producing text. It's more accurate, more nuanced, and better at maintaining context across long, complex documents.
👁️ Vision and Multimodal
Fable 5 also brings improvements to image and document understanding. If you upload a PDF, a chart, or a screenshot, Fable 5 is better at extracting meaning from it than any previous Claude model.
The Honest Part: Where Fable 5 Has Limits
Every honest review has to say this clearly.
The cybersecurity fallback is real. When you ask Fable 5 about offensive security topics, exploit techniques, or dangerous biology — it doesn't use Mythos-level intelligence. It falls back to Opus 4.8. For most users this never matters. For security researchers, it's worth knowing.
The price is higher. Fable 5 costs $10 per million input tokens and $50 per million output tokens. That's double Opus 4.8's price. Google is buying the volume market: $2 input is unmatched, and batch mode halves it again. OpenAI holds the middle with GPT-5.5, then charges a steep premium for GPT-5.5 Pro. Anthropic priced Fable 5 at double its own Opus 4.8 but well under OpenAI's top tier.
For API developers at high volume, the cost difference is real. For individual subscribers using the Claude app, you don't pay per token — you just use it.
The free window closes June 22. This is the most important thing for Claude subscribers to know right now.
How to Access Claude Fable 5 — Right Now
Here's the practical guide based on which plan you're on:
Claude Pro / Max / Team / Enterprise subscribers: Fable 5 is available right now at no extra cost — but only until June 22, 2026. After that, using it will require usage credits. Go to claude.ai, select a new conversation, and choose "Claude Fable 5" from the model selector. Use it as much as you can before the 22nd.
API / Developer users:
Claude Fable 5 is generally available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry. Model ID: claude-fable-5. Pricing: $10/$50 per million tokens. Batch API available at half price for async workloads.
Free plan users: Fable 5 is not available on the free tier right now. Sonnet 4.6 remains the free tier model.
Claude Fable 5 vs Opus 4.8 — Should You Switch?
The short answer: yes, if you're on a paid plan. Here's the longer answer.
For writing, analysis, research, and document work — Fable 5 is noticeably better. The output is sharper, requires less editing, and handles complex instructions more reliably.
For coding — if you use Claude Code or write code with Claude's help, the difference is significant. Fable 5 handles multi-file tasks and longer coding sessions dramatically better than Opus 4.8.
For quick, simple tasks — Fable 5 and Opus 4.8 are similar enough that you might not notice the difference. If you're asking "what's the capital of France," it doesn't matter which model you use.
For API cost-sensitive workloads — stick with Opus 4.8 at $5/$25 if your tasks don't require frontier-level performance. The price difference compounds at scale.
What About Claude Mythos 5?
The elephant in the room: Fable 5 is great, but Mythos 5 is still out there — and still better.
Mythos 5 will be deployed through the company's Project Glasswing as an upgrade to Mythos Preview. The company claims this new model has the strongest cybersecurity capabilities of any model.
Mythos 5 has the same underlying model as Fable 5, but with the safety classifiers removed for specific high-risk domains. Security teams at AWS, Apple, Microsoft, Google, Cisco, and CrowdStrike are using it to find vulnerabilities in critical software infrastructure — the kind of bugs that affect every device and every user on the planet.
You won't get access to Mythos 5 unless you're an approved Project Glasswing partner. And for good reason: Anthropic's Claude Mythos Preview has autonomously identified thousands of zero-day vulnerabilities across every major operating system and every major web browser.
Fable 5 is what happens when you take that capability and make it safe enough for everyone. It's a genuinely impressive engineering achievement — and the right call.
The Bigger Picture: Why This Release Matters
Claude Fable 5 isn't just a new model. It's a signal about where AI is going.
For months, the narrative was that the most powerful AI models were too dangerous to release publicly. Anthropic built something powerful enough to worry them — and then spent months building the safety systems to make it available anyway.
That's a different approach than just shipping the most capable thing possible and hoping for the best. Whether you think Anthropic's safety approach is the right one or too cautious, the outcome is real: a model that was locked away for months is now available to anyone with a $20/month subscription.
Fable 5's capabilities exceed those of any model Anthropic has ever made generally available.
And if the Stripe migration story is representative — a 50-million-line codebase in one day — the gap between what AI could do six months ago and what it can do today just got a lot wider.
Bottom Line
Claude Fable 5 is the most significant AI release of 2026 — not because of the benchmarks, but because of what it represents. The most powerful publicly available AI in the world just got dramatically more capable, and it's available to anyone on a paid Claude plan right now.
If you're a Claude Pro or Max subscriber: open claude.ai today, switch to Fable 5, and try it on something real. You have until June 22 before it moves to credits. That's 10 days of access to something genuinely different.
Don't waste it.
Follow Ampick for every major AI release explained clearly — ampick.xyz
0 Comments