Claude Fable 5 vs Opus 4.8: Which Model Should You Pick?

Anthropic shipped Claude Fable 5 on June 9, 2026, and it's a big deal for one specific reason: it's the first Mythos-class model anyone outside the company can actually use. That makes it the most capable Claude that's generally available. The catch is that it costs twice as much as Opus 4.8, which until last week was the top of the lineup and is still very, very good.

So the real question isn't whether Fable 5 is better. It is. The question is whether it's better enough, for the work you do, to justify paying double. We're independent here and not affiliated with Anthropic, so we'll try to give you the fair version rather than the launch-day hype. Short answer up front: for hard, long tasks Fable earns its price, and for most everyday work Opus 4.8 is still the smarter buy.

What actually changed with Fable 5

Fable 5 exceeds Opus 4.8 on nearly all benchmarks. The biggest gaps show up in three areas: long horizon complex tasks, finance, and coding. Those are exactly the jobs where a model has to hold a lot of context, reason through many steps, and not lose the thread halfway through.

A few concrete signals from the launch, since vague "it's smarter" claims aren't worth much:

• Stripe used Fable 5 to finish a 50 million line Ruby migration in a single day. That's the kind of work that would normally take a team roughly two months.
• It took the top score on Cognition's FrontierCode coding eval at medium effort.
• It came out on top on Hebbia's Finance Benchmark.
• Its vision is state of the art, good enough to rebuild working web apps directly from screenshots.

We're deliberately not quoting a context window, exact benchmark percentages, or a feature list beyond this, because Anthropic didn't publish those in a form we'd trust to repeat. If you see a blog throwing around precise numbers, treat them with suspicion. If you want the deeper breakdown of what the model can do, we wrote that up separately in our Claude Fable 5 explainer.

Where Opus 4.8 still holds up

Here's the part the launch posts gloss over. Opus 4.8 didn't get worse. It's a strong model that handles the large majority of real tasks without breaking a sweat: drafting, summarizing, normal coding, research, customer support, the everyday stuff. For most people, most of the time, the difference between the two on a given prompt is going to be small or invisible. The gap widens on the hard, long, high-stakes jobs, not on the routine ones.

The price difference is the whole story

Let's put the numbers next to each other, because this is where the decision actually lives.

• Claude Fable 5: 10 dollars per million input tokens, 50 dollars per million output tokens.
• Claude Opus 4.8: half of that, so 5 dollars per million input and 25 dollars per million output.

Double the price is not a rounding error. If you're running a high-volume pipeline, a chatbot fielding thousands of conversations, or any workload where token spend is a real line item, that 2x multiplier compounds fast. We dug into the math on different usage patterns in the Fable 5 pricing breakdown, but the short version is simple: the value depends entirely on whether your tasks are the kind Fable is meaningfully better at.

A useful way to think about it: you're not paying double for double the quality across the board. You're paying double for a noticeable edge on the hardest slice of your work. If most of your prompts aren't in that slice, you're overpaying.

The safety fallback you should know about

This part is easy to miss and it matters for how you budget and how you reason about reliability.

Fable 5 ships with classifiers that hand certain requests back to Opus 4.8. The categories are cybersecurity, biology and chemistry, and distillation. When one of those triggers, your prompt is effectively answered by Opus 4.8 instead of Fable, even though you called Fable.

Anthropic says this fires in under 5 percent of sessions on average. So for most users it almost never happens. But if your work lives in one of those domains, a security research team, say, or anything bio or chem adjacent, you may hit it more often than the average. The practical upshot: in a small share of cases you're paying the Fable price and getting an Opus 4.8 answer. That's not a scandal, it's a deliberate safety design, but you should know it exists before you assume every Fable call gives you the top-tier model.

Where you can get each one

Availability is straightforward.

Fable 5 is on the Claude API as model id 'claude-fable-5', plus Enterprise and AWS Bedrock. It's free inside the Pro, Max, Team, and seat Enterprise plans from June 9 to June 22, 2026, after which it moves to usage credits. So if you're on one of those plans, the smart move is to try Fable heavily during the free window and form your own opinion before the meter turns on.

Opus 4.8 is widely available, cheaper, and it's the model Fable falls back to anyway for the guarded topics. It's a safe default that you can run everywhere without thinking twice about cost.

When to pick each one

After all that, here's how we'd actually choose.

Reach for Fable 5 when:

• You're doing long horizon, multi-step work where losing the thread is expensive.
• You're shipping serious code, large migrations, or anything where one strong pass saves days of cleanup.
• You're working in finance or another domain where its benchmark lead is real.
• You need top-tier vision, like rebuilding a UI from a screenshot.
• You're inside the free window and might as well use the best.

Stick with Opus 4.8 when:

• Cost matters and you're running real volume.
• Your tasks are everyday quality work where the two are close anyway.
• You want the broadest availability with the fewest surprises.
• You're price sensitive and the marginal quality bump won't change your outcome.

There's no shame in defaulting to Opus 4.8 and pulling in Fable 5 only for the genuinely hard jobs. Honestly, that mixed approach is what we'd recommend to most teams. Use the cheap, excellent model for the bulk of the work and spend the premium where it actually moves the needle.

If you're also weighing Fable against the broader Mythos line, we compared those directly in Fable 5 vs Mythos 5.

Frequently asked questions

Is Claude Fable 5 better than Opus 4.8?

Yes, on nearly all benchmarks. The biggest gains are in long horizon complex tasks, finance, and coding. But "better on benchmarks" and "better for your specific work" aren't the same thing. On routine tasks the two are often close, and Opus 4.8 is still an excellent model.

Is Fable 5 worth double the price?

It depends on your work. Fable costs 10 dollars per million input tokens and 50 per million output, twice Opus 4.8. If you're tackling hard, long, high-stakes tasks where one strong pass saves real time, the premium pays off. If you're running everyday work at volume, the 2x cost usually isn't justified.

Which should I use?

Use Opus 4.8 as your default for cost and breadth, and bring in Fable 5 for the genuinely difficult jobs: large migrations, complex multi-step reasoning, finance, and serious vision work. That mix gets you most of Fable's upside without paying double for everything.

Does Fable 5 use Opus for some answers?

Yes. Fable has classifiers that hand cybersecurity, biology and chemistry, and distillation requests back to Opus 4.8. This triggers in under 5 percent of sessions on average, so in a small share of cases you're effectively getting an Opus 4.8 answer even though you called Fable.