Anthropic launched two new models on June 9, 2026: Claude Fable 5 and Claude Mythos 5. These are the first widely discussed models in a new "Mythos class" tier that sits above Claude Opus, which until now was Anthropic's most capable line.
The short version: Fable 5 and Mythos 5 are the same underlying model. Fable 5 is the public version, with safety filters built in.
Mythos 5 is the same model with some filters lifted, available only to a small group of approved organizations. The safeguards are the only thing separating them, which is also why they have different names. Fable comes from the Latin fabula, meaning "that which is told," which is related to the Greek word mythos.
Here's a full breakdown of what they do, the benchmark numbers, the real-world demos, the science, the pricing, and the catches you should know about.
Quick background: why "Mythos" matters
In April 2026, Anthropic released its first Mythos-class model, Claude Mythos Preview, but only to a limited group through a program called Project Glasswing.
The model turned out to be unusually good at finding and exploiting software vulnerabilities across major operating systems and browsers, which is exactly the kind of capability that is dangerous in the wrong hands. So Anthropic held it back and gave access only to vetted cyber defenders and infrastructure providers, working with the US government.
At the time, Anthropic said its goal was to eventually release Mythos-level capability to everyone, once it had safeguards strong enough to prevent misuse. Fable 5 is that release. It is the same tier of capability, wrapped in new safety systems so it can go public.
What is Claude Fable 5?
Claude Fable 5 is now Anthropic's most capable generally available model.
Anthropic says its capabilities exceed any model it has ever released publicly, and that it is state of the art on nearly every benchmark tested, with strong performance in software engineering, knowledge work, vision, and scientific research. The longer and more complex the task, the bigger its lead over older Claude models.
The core theme is autonomy. Fable 5 can work on its own for longer than any previous Claude model, which is what unlocks the bigger jobs below.
What Fable 5 can actually do
Software engineering
During early testing, Stripe reported that Fable 5 compressed months of engineering work into days.
On a 50 million line Ruby codebase, it performed a codebase-wide migration in a single day that would have taken a whole team over two months by hand. It is also more token-efficient than past models, scoring highest among frontier models on Cognition's FrontierCode evaluation even at medium effort.
Knowledge work
On Hebbia's finance benchmark for senior-level reasoning, Fable 5 posted the highest score of any model, with gains in document reasoning and chart and table interpretation. The trading firm IMC said it aced their trading-analysis tests nearly across the board.
Vision
This is one of the bigger leaps. Fable 5 can pull precise numbers out of detailed scientific figures and rebuild a web app's source code from screenshots alone. It also needs far less hand-holding: earlier Claude models struggled to play Pokémon FireRed even with helper tools, but Fable 5 finished the game using nothing but raw screenshots.
Memory and long context
Fable 5 stays focused across millions of tokens and improves its work using its own notes. Playing the deck-building game Slay the Spire, giving it file-based memory improved its performance three times more than it did for Opus 4.8, and it reached the final act three times more often.
The demos people are talking about
Anthropic showed off several tasks Fable 5 completed largely on its own, and they are worth seeing if you want a feel for the model:
It built a solar system simulation that derives the planets' orbital motion from physics first principles and uses it to predict solar eclipses.
It autonomously plays Factorio, the factory-building game, strategizing and building an automated factory by itself.
It designed a 3D-printable CAD model in a browser-based editor, and it also built that editor, including its AI copilot.
It coded a fluid simulation synced to the beat of a classical EDM remix, and it produced the music itself through code, despite never having heard music.
What is Claude Mythos 5?
Mythos 5 is the same model with the cyber safeguards lifted. It is the upgrade to Mythos Preview, and everyone who had Preview access (including Project Glasswing partners) is being moved to it. Anthropic says it has the strongest cybersecurity capabilities of any model in the world.
It is also where the most striking science results come from:
Drug design. Anthropic's internal protein design experts used Mythos 5 to speed up parts of the drug design process by around ten times. With the right tools but no human help, the model matched or beat skilled human operators, choosing binding sites, running design tools, and recovering from its own failures. Nine of 14 protein targets it worked on yielded strong drug candidates now under investigation.
Novel biology hypotheses. In blind comparisons, Anthropic's scientists preferred Mythos 5's molecular biology hypotheses about 80% of the time. One hypothesis, a new mechanism for an E. coli protein, was independently confirmed by another lab working on the same problem.
Genomics. Over about a week of mostly autonomous work, Mythos 5 assembled single-cell data for millions of cells across 138 animal species and trained its own machine learning model. With only high-level human input, that model outperformed a recent model published in the journal Science, despite being 100 times smaller.
Access to Mythos 5 still runs through Project Glasswing approval, in consultation with the US government, with plans to expand it through a broader trusted access program.
The benchmark numbers
Independent testing firm Artificial Analysis ranked Fable 5 number one on its Intelligence Index with a score of 64.9, about 5 points ahead of the closest non-Anthropic model, GPT 5.5. Anthropic now holds the top two spots, and Fable set the highest score on 5 of the 10 underlying benchmarks.
It also set a record on GDPval AA, a benchmark for real-world work tasks, scoring 1932 versus 1890 for Opus 4.8, and beat the previous knowledge leader, Gemini 3.1 Pro, by 7 points on the AA Omniscience accuracy and hallucination test.
The pattern is consistent: the biggest gains are in coding and long-running agent work, which is where Anthropic was already strongest.
How the safety system works
This is the genuinely new part, and worth understanding before you build on it.
Fable 5 runs separate AI systems called classifiers that watch for misuse in three areas: cybersecurity, biology and chemistry, and distillation. When a request trips one of these, Fable 5 does not answer. The request automatically falls back to Claude Opus 4.8, and the user is told when this happens. Anthropic deliberately tuned these classifiers to be cautious, so they will sometimes catch harmless requests too.
The reassuring number: Anthropic's early data shows more than 95% of Fable sessions involve no fallback at all, and in those sessions Fable performs effectively the same as Mythos 5. A fallback to Opus 4.8 is also a real answer from a strong model, not a flat refusal.
Here is what each classifier covers:
1. Cybersecurity: Mythos-class models are very good at finding and exploiting software vulnerabilities and at multi-step "agentic" hacking. The classifiers are built to block both, and Anthropic says they stop Fable from making any progress on these tasks.
2. Biology and chemistry: Anthropic used to block only a narrow set of bioweapons queries, but decided that was no longer enough given how much better models have gotten at real scientific tasks. For now, Fable falls back to Opus 4.8 on most biology and chemistry requests, a deliberately broad net they plan to narrow over time.
3. Distillation: This targets large-scale attempts to copy Claude's capabilities to train competing models, which could spread near-frontier ability without safeguards.
On jailbreak resistance, Anthropic ran an external bug bounty that found no universal jailbreak in over 1,000 hours of testing. It is being transparent that the UK AI Safety Institute made some progress toward one in early testing, while noting that fully preventing jailbreaks is likely impossible and the goal is to make them slow and costly enough to catch first.
Pricing and specs
Input: $10 per million tokens
Output: $50 per million tokens
Context window: 1 million tokens by default
Max output: 128k tokens per request
That is premium pricing, but it is less than half what Mythos Preview cost. It suits hard, high-value tasks rather than routine ones.
One technical note for developers: thinking is always on for these models. You cannot disable it, and the raw chain of thought is never shown. You control thinking depth through an effort parameter instead.
The data retention catch
Before you adopt it, know this. Fable 5 and Mythos 5 are "Covered Models," which means Anthropic requires 30-day retention of inputs and outputs for all traffic on Mythos-class models, on both Anthropic's own surfaces and third parties like AWS. Zero data retention is not available for these models.
Anthropic says it will not use this data to train models or for any non-safety purpose, that it logs all human access, and that it deletes the data after 30 days in almost all cases.
The point of keeping it is to catch complex attacks that span many requests and to reduce false positives. On AWS Bedrock, you have to explicitly opt into data sharing before you can even call the model, and that data leaves the AWS security boundary.
If your company has strict data residency or privacy requirements, factor this in before switching.
Where and when you can use it
For developers, Fable 5 is available now on:
The Claude API (model ID
claude-fable-5)Claude Platform on AWS
Amazon Bedrock (US East and Europe Stockholm regions at launch)
Google Vertex AI
Microsoft Foundry
On the API and consumption-based Enterprise plans, it is fully available immediately.
For subscription users, there is a deadline worth noting. Because demand is expected to be very high and hard to predict, Anthropic is rolling out in stages:
From launch through June 22, 2026, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost.
On June 23, it comes off those plans, and using it will require usage credits.
Once Anthropic has enough capacity, it plans to restore Fable 5 as a standard part of subscription plans.
If you want to try it on your existing plan for free, do it before June 22.
The bottom line
Fable 5 is the most capable model anyone can use right now, and the lead on coding and agentic work is real and independently confirmed.
The science coming out of Mythos 5, especially in drug design and genomics, hints at where this is heading. The tradeoffs are just as clear: premium pricing, mandatory 30-day data retention, a safety system that quietly routes some requests to a weaker model, and subscription access that pauses on June 22 until capacity catches up.
For teams doing serious software engineering, document analysis, or long-running automated work, it is worth testing now. For everyday tasks, cheaper models like Sonnet will still cover most needs at a fraction of the cost.
That renders as: If you want to go deeper, Anthropic published a full system card and risk report alongside the launch.