In This Article
Key Takeaways
- On June 9, 2026, Anthropic launched Fable 5 — the first generally available model in its new 'Mythos-class' tier, which sits above Opus.
- Fable 5 and Mythos 5 are the same underlying model; Fable 5 routes high-risk requests to Opus 4.8, while the less-restricted Mythos 5 is limited to vetted cyber defenders.
- Fable 5 scored 80.3% on SWE-bench Pro (vs. GPT-5.5's 58.6%) and even completed Pokémon FireRed from a vision-only harness.
- Both are priced at $10/$50 per million tokens — double Opus 4.8 — a premium that tells you they are for high-value, long-horizon work.
On June 9, 2026, Anthropic released Claude Fable 5 — the first model from its new "Mythos-class" tier, the level that now sits above the Opus class, cleared for general use. For builders who had just gotten comfortable with Opus 4.8, released only twelve days earlier, the ceiling moved again.
I use these tools every day, in teaching and in federal work, so let me lay out the new ladder clearly, explain the unusual two-name structure, walk through the genuinely striking benchmarks, and offer a calm way to think about a frontier that keeps rising. Because the right response to a more powerful model is rarely "switch everything to it."
What launched on June 9
Anthropic positioned Fable 5 as the most powerful generally available model it had ever shipped, built for long-horizon tasks — work that unfolds over many steps and a long stretch of time rather than a single quick answer. Mythos 5 sits in the same new top tier but is handled very differently, which is where the structure gets interesting.
A new tier above Opus
Until June 9, Opus was Anthropic's most capable tier. The Mythos class is a deliberate step above it. Think of it like a product line adding a new flagship: the old top model does not disappear or get worse, but there is now a more powerful — and more expensive — option for the heaviest jobs. Most everyday work still runs fine on the cheaper, faster models. The new tier exists for the tasks where the last increment of capability is worth real money.
Fable and Mythos: same model, different safeguards
Here is the part that confused a lot of early coverage. Fable 5 and Mythos 5 are the same underlying model. The difference is the safety wrapper around it. Fable 5 ships with classifier safeguards that detect and route high-risk requests — in cybersecurity, biology, chemistry, and model "distillation" — over to Opus 4.8 instead. That makes Fable 5 safe to release to the general public.
Mythos 5 is the same model with some of those safeguards lifted, made available only to a small, vetted group of cyber defenders and infrastructure providers. In other words: one engine, two doors. A guarded door for everyone, and a restricted door for trusted specialists who need the unfiltered capability for defensive work.
The benchmarks are striking
The numbers are not subtle. On SWE-bench Pro, a test of difficult engineering tasks, Anthropic reports Fable 5 scoring 80.3%, against 58.6% for OpenAI's GPT-5.5 — a wide gap on a hard benchmark. On a difficult biology reasoning test it scored 46.1%, ahead of Opus 4.8's 40.0%.
The qualitative demonstrations are even more vivid. Fable 5 completed Pokémon FireRed from a minimal, vision-only harness — earlier Claude models needed extra scaffolding just to make progress. It can reconstruct a web app's source code from screenshots and extract precise values from scientific figures. Mythos 5 matched or exceeded skilled human operators on protein-design tasks across 9 of 14 targets. These are the kinds of results that mark a real capability jump, not a marketing point.
Claude tiers after June 9, 2026
| Model | Tier | Input / Output ($/M) | Access |
|---|---|---|---|
| Fable 5 | Mythos-class | $10 / $50 | General (high-risk routed to Opus 4.8) |
| Mythos 5 | Mythos-class | $10 / $50 | Vetted cyber defenders only |
| Opus 4.8 | Opus-class | $5 / $25 | General |
| — SWE-bench Pro — | Fable 5: 80.3% | GPT-5.5: 58.6% | Higher is better |
The price tells a story
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens — double Opus 4.8's $5/$25, though Anthropic notes it is less than half the price of the earlier Mythos Preview. A "token" is roughly a chunk of a word; "input" is what you send the model and "output" is what it writes back, which costs more because generating is harder than reading.
Anthropic included Fable 5 at no extra cost on Pro, Max, Team, and Enterprise plans — but only from June 9 through June 22, 2026. After June 23, continued use requires usage credits. The doubled price is not arbitrary. It tells you who the model is for.
A new top tier does not make the old models worse. It just gives the very hardest problems a more powerful tool — at a price that says “use me only when it counts.”
Match the model to the task
You would not charter a private jet to cross town, and you would not use a Mythos-tier model to draft a quick email. The premium price is a signal: this tier is for the long, complex, valuable task where being meaningfully better is worth the cost.
For most learning and most everyday work, a fast, inexpensive model is the right choice — and often indistinguishable in quality for the task at hand. The discipline of matching the tool to the job is worth more than always reaching for the most powerful option.
The right question is never “which is most powerful?”
Every time a new flagship launches, the instinct is to switch to it. Resist that. The right question is always “what is the cheapest model that does my job well?” For drafting, summarizing, and learning, that is usually a fast mid-tier model. Reserve the premium tier for the genuinely hard work that justifies double the price. This single discipline will save you money and keep your attention on the work rather than on the leaderboard.
What it means for you
The frontier rising is not a reason for anxiety. It is a reason for clarity. Fable 5 is a remarkable model, and most people reading this will rarely need it. That is not a failure — it is the normal, healthy state of a maturing tool. Cars can go 150 miles an hour; you drive 65.
Learn to match the tool to the task. Reach for the powerful, pricey tier only when the work genuinely demands it. The model names will keep changing — there will be a Mythos 6 and a Fable 6 before long. The skill of choosing wisely will not go out of date. Build that skill, and the endless release cycle becomes something you watch with interest rather than anxiety.
Learn to choose and use the right AI model
Our bootcamp teaches you how to match models to tasks, control costs, and build real products — not chase every leaderboard. Five U.S. cities, June through October 2026.
See Our BootcampSources: VentureBeat, “Anthropic brings Mythos to the masses with Claude Fable 5”; Vellum and Finout benchmark and pricing breakdowns; claudefast and CloudInsight guides to Fable 5 / Mythos 5. Figures (80.3% SWE-bench Pro, $10/$50 pricing, June 9–22 free window, same-model/different-safeguards structure) reflect Anthropic's published statements and independent coverage.
The capabilities behind the benchmarks
Benchmark numbers can feel abstract, so consider what Fable 5 actually demonstrated. It completed Pokémon FireRed from a minimal, vision-only setup — meaning it played a video game by looking at the screen and deciding what to do, with almost no special scaffolding, something earlier Claude models could not manage unaided. It reconstructed a web application's source code from nothing but screenshots. It extracted precise numerical values from scientific figures. Mythos 5, the less-restricted sibling, matched or beat skilled human operators on protein-design tasks across 9 of 14 targets.
These are not parlor tricks. Each one points at a real, general capability: perceiving an environment and acting in it, reverse-engineering a system from its appearance, reading data out of images, and reasoning at expert level in a hard scientific domain. When a model can do all of these, the benchmark scores stop being the story and start being the evidence for something broader — a model that can take on genuinely open-ended, long-horizon work.
Why the two-door safety design matters
The split between Fable 5 and Mythos 5 is worth dwelling on, because it represents a thoughtful answer to a genuine dilemma. The same capability that lets a model help a cyber defender find vulnerabilities could help a cyber attacker exploit them. The same reasoning that aids legitimate biology research could, in the wrong hands, be misused.
Anthropic's answer was to ship one model behind two doors. The public door — Fable 5 — automatically routes high-risk requests in cybersecurity, biology, and chemistry to the more constrained Opus 4.8, so the average user gets the model's general power without the most dangerous edges. The restricted door — Mythos 5 — opens those edges only to a small, vetted group of defenders and infrastructure providers who need them for protective work. It is an imperfect solution to an unavoidable tension, but it is a serious one, and it reflects the kind of careful trade-off that responsible AI deployment increasingly requires. The export-control suspension that followed days later — which I cover in a companion piece — shows just how fraught that territory is.
Common questions
What's the difference between Fable 5 and Mythos 5? They are the same underlying model. Fable 5 ships with classifier safeguards that route high-risk cybersecurity, biology, and chemistry requests to Opus 4.8, making it safe for general use. Mythos 5 has some safeguards lifted and is available only to a small group of vetted cyber defenders and infrastructure providers.
How much do they cost? Both are $10 per million input tokens and $50 per million output tokens — double Opus 4.8's $5/$25. Anthropic included Fable 5 at no extra cost on Pro, Max, Team, and Enterprise plans only from June 9 through June 22, 2026; after that, continued use requires usage credits.
Is Fable 5 actually better than Opus 4.8? On hard benchmarks, meaningfully so — 80.3% on SWE-bench Pro versus Opus 4.8's lower score and GPT-5.5's 58.6%. It can reconstruct web-app source code from a screenshot and even completed Pokémon FireRed from a minimal vision-only setup. But 'better' depends on the task; for everyday work a cheaper model is usually the right call.
Should I switch everything to Fable 5? No. Use the cheapest model that does your job well. Reserve a premium, double-priced tier like Fable 5 for genuinely hard, high-value, long-horizon tasks. Matching the tool to the task saves money and keeps you focused on the work itself.