Anthropic's Claude Fable 5 Drops — and the AI Industry Is Buzzing About Its Coding Skills

Anthropic just dropped Claude Fable 5, and if you've been following the AI space at all, you'll want to pay attention — because this one feels different.

Here's the deal: Fable 5 is Anthropic's most powerful AI model ever made available to the public. And the headline capability everyone's buzzing about is coding. Not just "write me a function" coding — we're talking about the kind of deep, complex software work that normally takes a whole engineering team weeks or months.

The most striking real-world example came from Stripe, the payment processing giant. <a>During early testing, Stripe reported that Fable 5 compressed months of engineering into days.</a> Specifically, <a>in a 50-million-line Ruby codebase, the model performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand.</a> That's not a party trick — that's the kind of output that makes CTOs sit up straight.

On paper, the benchmark numbers back it up. <a>On SWE-Bench Pro, Claude Fable 5 scored 80.3%, significantly ahead of GPT-5.5's 58.6% and Gemini 3.1 Pro's 54.2%.</a> On a harder coding benchmark called FrontierCode Diamond, <a>Claude Fable 5 achieved 29.3%, compared with just 5.7% for GPT-5.5.</a> That's not a close race.

But coding isn't the whole story. <a>Fable 5 is the new state-of-the-art model for tasks involving vision</a> too — it can rebuild a web app's source code from screenshots alone. And to show off that vision capability in a very nerdy, very relatable way, Anthropic shared a timelapse of Fable 5 beating Pokémon FireRed using only raw game screenshots — no cheat codes, no maps, no extra tools. <a>Earlier Claude models needed a complex helper harness to play Pokémon; Claude Fable 5 completed the game with vision alone.</a>

What makes this launch technically significant is what Fable 5 actually is under the hood. <a>Mythos-class models are a tier of Claude models that sit above the Opus class in capability.</a> Until now, Anthropic kept these powerful Mythos-class models locked away — first revealed earlier this year through a restricted government program called Project Glasswing for cybersecurity experts. Fable 5 is, essentially, that same powerful engine with safety guardrails bolted on for the public.

And those guardrails matter. <a>The model's capabilities in specific areas like cybersecurity, biology, and chemistry are advanced enough that they could be misused to create wide-reaching cyberattacks or build dangerous bioweapons.</a> So Anthropic built in automatic rerouting: <a>if a user's request enters one of these high-risk areas, Fable 5 may automatically switch the task to Anthropic's older Claude Opus 4.8 model,</a> and the company says this happens in fewer than 5% of sessions.

For regular users on paid plans (Pro, Max, Team), Fable 5 is available now at no extra cost through June 22. After that, you'll need usage credits to access it. <a>Down the line, Anthropic plans to fold the model back into regular subscription plans once it has enough capacity.</a>

For developers and businesses using the API, pricing is $10 per million input tokens and $50 per million output tokens — <a>less than half the price of Claude Mythos Preview, but still roughly twice the price of Claude Opus 4.8.</a>

Why does this matter to you personally? If you're a developer, designer, or knowledge worker — someone who uses AI to help build, research, or analyze — this is a meaningful step up. The model isn't just faster at answering questions; <a>Fable 5 and Mythos 5 can work autonomously for longer than any previous Claude models,</a> which means it can actually take on multi-step projects rather than just single prompts. Think of it less as a smarter search engine and more as a capable colleague who can own a task from start to finish.