Build Apps with AI (No-Code)

Ship a real, working web app without writing code by hand. You'll describe what you want in plain English to tools like Lovable, Bolt, and Cursor, then add a database, accounts, payments, and deploy it live.

View course →

Claude Fable 5 and Mythos 5: Everything Anthropic's New Frontier Model Can Do

On June 9, 2026, Anthropic released Claude Fable 5 and Claude Mythos 5 — two models that share the same underlying brain but ship with different safeguard configurations. Together they introduce a new "Mythos class" that sits a tier above the Opus class we have spent the last year building on. If you work with AI for a living, or you are learning to, this is the most important release of the season. Below is a clear, practical breakdown of everything it can do, why it matters, and how to put it to work.

Two models, one engine

Let's get the naming out of the way first, because it confuses people.

• Claude Fable 5 is the general-availability model. It is available everywhere today through the Claude API and is rolling into the subscription plans (Pro, Max, Team, Enterprise).
• Claude Mythos 5 is the same model with certain safeguards lifted, released only to vetted partners — research labs and organizations in Anthropic's trusted-access programs.

In other words, you and I will use Fable 5. Mythos 5 is the version that select scientists and security teams get under supervision. When people quote jaw-dropping benchmark results from "Mythos," remember it is the same intelligence you are getting in Fable, just with a different set of guardrails around the edges.

State of the art across the board

Anthropic claims Fable 5 reaches state-of-the-art results on nearly every benchmark it was tested on, with particular strength in software engineering, knowledge work, vision, and scientific research. The headline shift is not raw IQ on a single test — it is endurance. Fable 5 can work autonomously for longer stretches than any previous Claude, staying coherent on long, multi-step tasks where earlier models drifted or lost the thread.

That endurance is the thing to internalize. The previous generation was great at sprints: a function, an email, a summary. This generation is built for marathons: a migration, an investigation, a research project that unfolds over hours.

Software engineering: months compressed into days

The coding results are the most concrete. Stripe reported that Fable 5 "compressed months of engineering into days," including a 50-million-line Ruby codebase migration completed in a single day — work their team estimated would take more than two months by hand. On Cognition's FrontierCode evaluation, Fable 5 posted the highest score among frontier models, and it does this while being more token-efficient than past Claude models, meaning each unit of work costs less.

What this looks like in practice:

• Large refactors and migrations that used to be quarter-long projects become week-long ones.
• Reading an unfamiliar codebase and proposing correct changes across many files in one pass.
• Fewer turns to done. Early partners like Ramp noted Fable 5 completes tasks in fewer back-and-forth turns, and Replit said it "understands what builders mean, not just what they type."

For anyone learning to build with AI, the lesson is that the bottleneck is moving from "can the model write code?" to "can you describe the outcome clearly and supervise the work?" That is a skill you can train.

Knowledge work: senior-level reasoning over documents

Fable 5 posted the highest score on Hebbia's Finance Benchmark for senior-level reasoning, and Hebbia called it the first model to break 90% on their core analytics benchmark — a 10-point jump over Opus 4.8. The trading firm IMC said it "aced their trading-analysis evaluations nearly across the board."

Translated to everyday work, this is the model you hand a messy quarterly report, a 200-page contract, or a folder of spreadsheets and ask for an answer you can actually defend. It reads charts and tables, reasons across documents, and holds the details together instead of hallucinating a tidy-but-wrong summary.

Vision: it can see, and act on what it sees

Vision is where the demos get fun. Fable 5 is state-of-the-art on vision tasks and can:

• Extract precise numbers from scientific figures — not "this chart trends up" but the actual values.
• Rebuild a web app's source code from a screenshot alone. Show it a UI, get working code back.
• Complete Pokémon FireRed using vision only, where previous Claude models needed helper tools to interpret the screen.

For builders, "screenshot to code" is a genuine workflow unlock. For analysts and researchers, "read the figure accurately" removes one of the most error-prone steps in working with real-world documents.

Memory and long context: focus across millions of tokens

This is the quiet superpower. Fable 5 maintains focus across millions of tokens and uses a persistent, file-based memory to improve its own outputs over time. Anthropic's most vivid example: in Slay the Spire game testing, file-based memory improved performance 3x over Opus 4.8, and the model reached the final act three times more often when it could write and read its own notes.

Why should you care about a card game? Because it proves the model can run a long task, notice what worked, write that down, and use it later — the loop that separates a clever autocomplete from a genuine collaborator. For long projects, give Fable 5 a place to keep notes and it gets meaningfully better.

Science: novel hypotheses, drug design, and genomics

The Mythos 5 results in the sciences are the most striking part of the announcement:

• In drug design, internal experts said Mythos 5 accelerated parts of their work around tenfold, matching or beating skilled human operators on certain protein-design tasks and generating 9 of 14 protein targets that yielded strong drug-design candidates.
• It is the first Claude model to consistently produce novel, compelling scientific hypotheses — scientists preferred its molecular-biology hypotheses about 80% of the time in blind comparisons, and one hypothesis about an E. coli protein was later corroborated by independent lab work.
• In a one-week autonomous genomics run, it assembled single-cell data for millions of cells across 138 animal species and built a custom ML model that outperformed a recent Science journal publication while being 100x smaller.

You will not be designing proteins this afternoon, but the signal matters: the model is crossing from "summarize what is known" into "propose something new that turns out to be right."

Pricing and availability

The economics improved sharply. Fable 5 is priced at $10 per million input tokens and $50 per million output tokens — described as less than half the price of the earlier Mythos Preview.

On subscriptions, the rollout has a calendar worth noting:

• June 9–22: Fable 5 is included in Pro, Max, Team, and seat-based Enterprise at no extra cost.
• From June 23: access uses usage credits.
• Once Anthropic has enough capacity, the plan is to restore it as a standard feature.

So there is a real window right now to use the best model Anthropic has ever shipped at no premium — if you know how to drive it.

Safety, briefly

Fable 5 ships with classifier-based safeguards in three areas: cybersecurity, biology/chemistry, and distillation (preventing competitors from copying its capabilities). The notable design choice: when a query trips a safeguard, it falls back to Claude Opus 4.8 instead of refusing — so you rarely hit a hard wall. Safeguards trigger in under 5% of sessions on average, and external red-teaming logged 1,000+ hours without finding a universal jailbreak. The point for normal users: the guardrails are mostly invisible.

How Fable 5 compares to the models you already use

If you have been working with Opus 4.8, the upgrade is not a personality change — it is a capability and stamina change. The same prompts you already write will generally produce better, more reliable results, and the tasks you avoided handing to AI because they were "too big" are now on the table. Three differences are worth feeling out for yourself:

• It quits less. Where older models would stop early, hand back a partial answer, or quietly drop requirements on a long task, Fable 5 keeps going and keeps tracking the original goal.
• It asks for clarity less and infers intent more. Replit's note that it understands "what builders mean, not just what they type" is the practical headline — vague prompts degrade more gracefully.
• It costs less per result. Higher token efficiency means the same job consumes fewer tokens, so even at frontier quality the effective price per finished task drops.

The mental model: treat Opus-class models as fast, sharp assistants for discrete tasks, and treat Fable 5 as the one you trust with a project.

How to start using Fable 5 today

You do not need to change your tools to benefit. A practical first week:

1. Re-run a task that disappointed you before. Take a prompt where an older model fell short — a long analysis, a multi-file code change — and run it again on Fable 5. Calibrating on your own work is worth more than any benchmark.
2. Give it room to think. Fable 5 rewards bigger, well-scoped requests over a chain of tiny ones. Describe the end state, the constraints, and what "done" looks like, then let it work.
3. Use the free window. From June 9–22 it is included in paid plans at no extra cost. After June 23 it draws on usage credits, so this is the moment to build the habit.
4. Keep a human checkpoint. More autonomy means more output to review. Decide upfront where you will verify before anything ships — code, numbers, or claims.

Quick answers

Is Fable 5 free? It is included in Pro, Max, Team, and Enterprise from June 9–22; after that, access uses usage credits until Anthropic restores it as a standard feature.

Should I use Fable or Mythos? You use Fable 5 — Mythos 5 is restricted to vetted partners. They are the same model with different safeguards.

Will my old prompts still work? Yes, and most will produce better results. The biggest gains come from giving it longer, more ambitious tasks.

Launching next week: our Pro mini-course on Claude Fable 5

A model this capable is only as good as your ability to direct it — and that is exactly what we teach. Next week, starting June 16, we are launching a brand-new 4-lesson mini-course on Claude Fable 5, exclusive to Pro subscribers.

Here is the curriculum:

1. Lesson 1 — Fable 5 fundamentals. Fable vs. Mythos, where it fits among the models you already use, and how to set up your workspace to get the most out of it.
2. Lesson 2 — Long-horizon work and memory. Structuring multi-hour tasks, giving the model file-based memory, and supervising autonomous runs without losing control.
3. Lesson 3 — Vision and coding workflows. Screenshot-to-code, reading real documents and figures, and running large refactors safely.
4. Lesson 4 — Knowledge-work playbooks. Repeatable prompts for analysis, research, and decision support you can drop into your day-one workflow.

Every lesson is hands-on, built around the kind of work you actually do, and designed so you finish able to ship something real.

Get the mini-course — go Pro

The Claude Fable 5 mini-course is included with Pro at no extra charge. Pro members also get every premium course, the full prompt library, and new drops like this one the moment they launch.

Subscribe to Pro for $10/month →

Join before June 16 and the mini-course lands in your account on day one.

The takeaway

Claude Fable 5 is not an incremental update — it is a new tier defined by stamina, vision, memory, and the early signs of genuine scientific creativity. The models keep getting better; the durable advantage is knowing how to use them. Subscribe to Pro, join the mini-course next week, and turn this release into a real skill instead of a headline you scrolled past.