Claude Fable 5 and Mythos 5 Released: Anthropic’s Biggest AI Update Yet

Two new Anthropic flagships just landed at the same time, and yes, the naming is confusing on purpose. Claude Fable 5 is the public version of a new “Mythos-class” model, and Claude Mythos 5 is the same underlying brain with the safety guardrails loosened for vetted security partners. Both went live on June 9, 2026, and both are state-of-the-art on almost every capability benchmark Anthropic has ever measured (Anthropic, June 9, 2026).

I spent the morning reading the launch post, the Mythos model page, and a stack of third-party coverage from TechCrunch, CNBC, and Business Insider. Here’s the plain-English breakdown of what shipped, what changed, and what it costs you.

Anthropic just released its most capable model ever — twice

Anthropic released two models on June 9, 2026 that share the same weights but ship with different safety stacks. Claude Mythos 5 is the raw model, reserved for vetted cybersecurity partners through Project Glasswing. Claude Fable 5 is the same model wrapped in classifier-based safety filters that automatically reroute sensitive queries to Claude Opus 4.8 (Anthropic, June 9, 2026).

Why the two names? Anthropic says “Fable” comes from the Latin fabula, meaning “that which is told.” The point is that the safeguards are what separate the two models, not the underlying capability (Anthropic blog post).

“Fable 5’s capabilities exceed those of any model we’ve ever made generally available. It is state-of-the-art on nearly all tested benchmarks of AI capability.”

— Anthropic, June 9, 2026

To be clear: Mythos is a new capability tier above Opus. The first Mythos model, Mythos Preview, shipped in April 2026 to about 50 critical-infrastructure partners. Fable 5 is the first Mythos-class model the public can touch, and Mythos 5 is the same model with the cyber filters lifted for those original Glasswing partners (Anthropic Glasswing update, May 22, 2026).

What changed since Claude Opus 4.8

If you remember Opus 4.8 (released May 28, 2026), Fable 5 is its bigger sibling, and you can feel it everywhere. Anthropic reports Fable 5 scores more than 10% higher than Opus 4.8 on certain internal capability benchmarks (CNBC, June 9, 2026). The longer and more complex the task, the bigger the gap.

Here are the wins I’m paying attention to:

• Software engineering: Stripe told Anthropic that Fable 5 performed a codebase-wide migration on a 50-million-line Ruby repo in a single day. A team of humans, by their estimate, would have needed over two months.
• Finance reasoning: On Hebbia’s Finance Benchmark for senior-level analysis, Fable 5 posted the highest score of any tested model.
• Vision: Fable 5 beat Pokémon FireRed using only raw screenshots. Previous Claude models needed a complex helper harness to even make progress on the game.
• Memory: In a Slay the Spire test with persistent file-based memory, Fable 5 improved its score 3x more than Opus 4.8 did with the same trick, and reached the final act three times more often.

The whole vibe is “agent that stays on task for hours, not minutes.” That’s the meaningful upgrade.

The benchmarks, side by side

Here’s how Anthropic positioned Fable 5 and Mythos 5 against the rest of the field on day one. I’m pulling these numbers straight from the official launch post, with the third-party SWE-bench Verified figure cross-checked against Morph’s running benchmark tracker.

Benchmark	Claude Fable 5 / Mythos 5	Claude Opus 4.8	Closest external rival
SWE-bench Verified	~95%	88.6%	GPT-5.5 (~86%)
Hebbia Finance Benchmark	Highest of any model	Second	Gemini 3.1 Pro
CursorBench	State of the art	Below	n/a
Cognition FrontierCode	Highest at medium effort	Below	n/a
Pokémon FireRed (vision-only)	Beats the game	Cannot complete	n/a

Third-party validation lines up with Anthropic’s claims. Stripe, Cursor, GitHub, Cognition, Hebbia, and IMC all submitted named quotes for the launch post, every one of them positioning Fable 5 above Opus 4.8 on a specific real-world test (Anthropic, June 9, 2026).

The safety story is the actual story

Here’s the part that matters more than benchmarks: Fable 5 ships with a multi-layer classifier system that decides, query by query, whether you get Mythos-grade intelligence or a polite fallback to Opus 4.8 (Anthropic, June 9, 2026).

The classifiers flag requests in three areas:

1. Cybersecurity — anything that smells like exploit development, reconnaissance, lateral movement, or offense. Anthropic says its classifiers prevented Fable from making any progress on internal cyber evaluations, and external red-teamers ran over 1,000 hours of jailbreak testing without finding a universal bypass.
2. Biology and chemistry — questions about dangerous pathogens, viral design, or bioweapons get rerouted, even though Mythos 5 itself excels at protein design (Anthropic AAV evaluation).
3. Distillation — large-scale attempts to mine Claude’s outputs to train rival models also fall back to Opus 4.8, a stance Anthropic has been tightening since earlier this year (Anthropic distillation policy).

In practice, fewer than 5% of Fable 5 sessions trigger a fallback, and more than 95% of sessions run on the Mythos model end-to-end (Anthropic, June 9, 2026). You’ll see a notice in the response when it happens.

Pull-quote: “More than 95% of Fable sessions involve no fallback at all — for those sessions, Fable 5’s performance is effectively the same as that of Mythos 5.” — Anthropic, June 9, 2026

There’s also a new data policy: Anthropic now requires 30-day retention on all traffic for Fable 5, Mythos 5, and any future model at this capability level, even for enterprise customers who previously had zero-retention contracts (Anthropic support article, TechCrunch coverage). Anthropic says it won’t train on it — only use it to catch jailbreaks and tune false positives.

Pricing: cheaper than Mythos Preview, pricier than Opus

Both new models price at $10 per million input tokens and $50 per million output tokens (Anthropic pricing page, confirmed by TechCrunch and CNBC, June 9, 2026).

Two reference points:

• Half the price of Mythos Preview, the previous top-tier Anthropic model (Anthropic launch post).
• 2x the price of Claude Opus 4.8, which still costs $5 input / $25 output (TechCrunch, June 9, 2026).

If you’re building on the API, your bill roughly doubles when you switch from Opus 4.8 to Fable 5. Whether that buys you enough extra capability depends on your workload, and Anthropic’s Dianne Penn told CNBC early customers are reporting better ROI per task even at the higher price.

How to actually get Fable 5 today

Availability depends on which Anthropic plan you’re on, and there’s a clock ticking. From the official announcement:

• Claude API and consumption-based Enterprise plans: Fable 5 is fully available right now via the model ID claude-fable-5.
• Pro, Max, Team, and seat-based Enterprise subscriptions: Fable 5 is included at no extra cost from June 9 through June 22, 2026. After that, it’s pulled and you’ll need to burn usage credits to keep using it.
• Mythos 5: Restricted to existing Glasswing partners and the upcoming biology trusted-access program.

Anthropic says it expects demand to spike hard, and capacity will be the limiting factor (Anthropic launch post). If you’re on a subscription plan and want the free window, you have about two weeks from today.

What it actually feels like to use

Pull-quotes from the launch post are usually marketing fluff, but the early-access customers here are unusually specific. A few that stood out to me:

• GitHub’s Mario Rodriguez said Fable 5 took on “complex, long-horizon coding tasks with a level of autonomy and reliability that exceeded previous benchmarks” (Anthropic, June 9, 2026).
• Rakuten’s Yusuke Kaji said: “At the highest effort, Claude Fable 5 reflects on and validates its own work. For us, that’s what makes highly autonomous operations possible — the extra thinking pays for itself.”
• Cognition’s Scott Wu reported Fable 5 is the highest-scoring model on FrontierBench and “excels at long-horizon reasoning.”
• Hex’s Izzy Miller said it’s the first to break 90% on Hex’s analytics benchmark, a 10-point jump over Opus 4.8.
• LushBinary’s analysis called it the first Mythos-class model made safe for general access, scoring “10%+ above Opus 4.8” on some benchmarks (LushBinary, June 9, 2026).

The pattern across every quote is the same word: autonomy. This isn’t a smarter autocomplete. It’s a model you can hand a multi-day task to and trust to come back with something finished.

Where Mythos 5 is already changing science

On the Mythos 5 side, the early research results are wild. Anthropic’s internal teams and biotech partners shared concrete numbers:

• Drug design: Mythos 5 accelerated parts of the drug-design process by roughly 10x. In a head-to-head on 14 protein targets, 9 yielded strong candidates that are now being investigated further (Anthropic, June 9, 2026).
• Molecular biology hypotheses: In blinded comparisons, Anthropic scientists preferred Mythos’s hypotheses over Opus-class models ~80% of the time. One Mythos hypothesis on an E. coli protein mechanism was independently corroborated in a separate biorxiv study.
• Genomics: Mythos 5 ran for over a week of largely autonomous work across 138 animal species, training a custom model 100x smaller than a recent Science-published model that it outperformed on cell-type identification.

These are not toy benchmarks. They’re real lab work.

How to choose: Fable 5 vs Opus 4.8 vs Mythos 5

Here’s my quick rule of thumb after reading the launch coverage and cross-checking with the Opus 4.8 page:

1. Pick Claude Fable 5 if you’re building on the API and want the best general intelligence Anthropic offers, with automatic safety fallback. Use the claude-fable-5 model ID.
2. Stick with Claude Opus 4.8 if your budget is tight, your tasks are short-form, or you’re in a regulated workflow that can’t tolerate any classifier rerouting. It’s still a top-3 frontier model as of late May 2026.
3. You cannot get Claude Mythos 5 unless you’re a Glasswing partner or a vetted biology researcher. There’s no consumer path right now, and Anthropic has not opened a public waitlist beyond the interest form.

If you want the public Mythos experience but can’t wait for the trusted-access program, Fable 5 is the answer. You get ~95% of the sessions running on the full Mythos model anyway (Anthropic, June 9, 2026).

What to watch next

A few things I’m keeping an eye on over the next two weeks:

• Subscription rollback on June 23: Pro, Max, Team, and seat-based Enterprise plans lose included Fable 5 access after this date unless capacity allows an extension (Anthropic, June 9, 2026).
• Biology trusted access program: Anthropic said it will let a small number of life-science researchers in for Fable 5 with the bio and chem filters removed, and cyber filters kept on (Anthropic, June 9, 2026).
• Wider Mythos access: Anthropic plans to expand Mythos 5 access through a “broader trusted access program” but has not given a date.
• Classifier tuning: Anthropic explicitly promised to reduce false-positive fallback rates as the safeguards get smarter. Watch the published <5% trigger rate move.

For now, the practical move is simple. If you’re already an Anthropic developer, point your toughest long-horizon task at claude-fable-5 and watch what happens. If you’re a subscription user, you have until June 22, 2026 to use it included before credits kick in. And if you’re just a curious observer, this is the first time a Mythos-tier model has been in reach of the public at all. That alone makes June 9, 2026 a real turning point for what Anthropic is willing to ship.