Claude Fable 5 Is Here and It Just Embarrassed Every Other Model

So Anthropic dropped a model yesterday that's powerful enough they had to build a bouncer into it. Let that sink in for a second.

Yesterday Anthropic made Claude Fable 5 generally available. It's their first "Mythos-class" model released to the public, which is a fancy way of saying it's the most capable thing they've ever let regular humans touch. Not a research preview. Not a waitlist. You can use it today.

Here's the 47-second version if you'd rather watch me say it with my hands:

What "most powerful model ever" actually means this time

Every AI company says "most powerful model ever" roughly every six weeks. (It's basically a sneeze at this point.) So let me show you the numbers instead of the marketing.

On SWE-bench Pro, the benchmark that measures whether a model can actually fix real software, Fable 5 hit 80.3%. Opus 4.8, which was already excellent, sits at 69.2%. That's an eleven-point jump. In benchmark land, eleven points is not an improvement. It's a different weight class.

Spatial reasoning nearly tripled (38.6% versus Opus 4.8's 14.5%). Legal reasoning went from "barely conscious" to category-leading. Replit clocked it as the top performer on their end-to-end vibe-coding test. One spreadsheet-automation company said it beat Opus at every setting while finishing 25 to 30 percent faster.

The pattern underneath all of it: the longer and uglier the task, the bigger Fable 5's lead gets. Short prompts, everybody looks smart. Hand it a sprawling multi-step mess and Fable 5 pulls away. Which, if you do real work, is the only thing you actually care about.

The longer and more complex the task, the larger Fable 5's lead. Short prompts make every model look smart. Real work is where this one separates.

The part nobody's talking about: it has a bouncer

Here's the genuinely interesting bit. Fable 5 is powerful enough that Anthropic shipped it with safety classifiers baked in for three areas: cybersecurity, biology and chemistry, and model distillation.

When you ask Fable 5 something that trips one of those wires, it doesn't refuse. It quietly hands your question to Claude Opus 4.8 instead, and tells you it did. Think of it as a brilliant specialist who, the second the conversation turns to weapons-grade chemistry, slides the question to the more careful colleague down the hall.

This happens in under 5% of sessions, and you don't get billed Fable rates when it does. So for basically everything you'll ever do (coding, research, writing, analysis) you get the full beast. The handoff only kicks in on the genuinely spicy stuff.

It's a weirdly elegant solution. Most companies handle dangerous capabilities by making the model dumber for everyone. Anthropic kept it sharp and just rerouted the 5% that matters.

What it costs you

Power has a price, and here it's literally double. Fable 5 runs $10 per million input tokens and $50 per million output. Opus 4.8 is $5 and $25. So you're paying 2x.

Is it worth it? Depends entirely on the job. (Shocking, I know.) For quick stuff, casual chat, or anything Opus already nails, no, don't burn the money. For long, gnarly, high-stakes tasks where the quality gap actually shows up and finishing 30% faster saves you real hours, yeah, the math works fast. Use the big expensive hammer on the big expensive nails. Use Opus for everything else.

The hype cycle will tell you Fable 5 changes everything. It doesn't. But it's the first publicly available model where "most powerful ever" isn't just a press release. Go run your hardest task through it and watch what happens.