Anthropic Released Fable 5, and It Cut Drug Design Time Tenfold

Welcome back to SavvyMonk, your one-stop for AI and tech news that actually matters.

Anthropic put out its most capable model ever today, and the part worth your attention is not the coding speed but what the system did inside a biology lab. There is also a free-access window that closes on June 22, so this one matters if you pay for Claude.

Let's get into it.

Last month, 48% of visitors to documentation sites across Mintlify were AI agents, not humans.

Claude Code, Cursor, and other coding agents are becoming the actual customers reading your docs. And they read everything.

This changes what good documentation means. Humans skim and forgive gaps. Agents methodically check every endpoint, read every guide, and compare you against alternatives with zero fatigue.

Your docs aren't just helping users anymore. They're your product's first interview with the machines deciding whether to recommend you.

That means: clear schema markup so agents can parse your content, real benchmarks instead of marketing fluff, open endpoints agents can actually test, and honest comparisons that emphasize strengths without hype.

Mintlify powers documentation for over 20,000 companies, reaching 100M+ people every year. We just raised a $45M Series B led by @a16z and @SalesforceVC to build the knowledge layer for the agent era.

Make Your Docs Agent-Ready

Anthropic released two models on June 9, and they share one underlying system. Fable 5 is the version built for general use, and it now sits at the top of nearly every capability benchmark the company tested, with the lead growing as tasks get longer and more complex.

The second model, Mythos 5, is that same system with its cybersecurity safeguards removed, and it is going to a narrow group of cyber defenders and infrastructure providers through Project Glasswing alongside the US government.

Both belong to what Anthropic calls its Mythos class, a tier that sits above the Opus models most people have been using until now. The naming carries the logic, since Fable comes from the Latin word for a story that is told and Mythos is its Greek cousin, and the safeguards are the only thing that separates the two.

The Engineering Story

The clearest signal of how far this jumps came from Stripe, which ran an early test on a Ruby codebase of 50 million lines and watched Fable 5 finish a migration across the whole thing in a single day, work that a full team would have spent more than two months doing by hand.

On Cognition's FrontierCode evaluation, which checks whether a model can pass hard coding tasks while holding to production-quality standards, Fable 5 scored highest among frontier models even when running at medium effort.

Michael Truell, the chief executive and co-founder of Cursor, said the model opened up a class of long-horizon problems that earlier systems could not reach, and Mario Rodriguez, the chief product officer at GitHub, pointed to a level of autonomy on complex coding work that went past previous benchmarks. The gains reach beyond code.

Fable 5 is now the strongest model for vision tasks, able to pull exact numbers out of dense scientific figures and rebuild a web app's source from screenshots alone, and it finished Pokemon FireRed using only raw game images after earlier Claude models needed a full helper harness to play at all. Given file-based memory in the deck-building game Slay the Spire, it improved three times more than Opus 4.8 did and reached the final act three times as often.

The Science

The headline result lives in the lab. Working with Mythos 5, Anthropic's internal protein design team sped up parts of the drug design process by roughly ten times, and on a set of 14 protein targets the model produced strong candidates on 9 of them while choosing binding sites, running design tools, and recovering from its own failures with no human stepping in. It pushed further into open science as well.

In blinded comparisons against Opus-class models, Anthropic's scientists preferred the molecular biology hypotheses that Mythos 5 wrote on its own about 80 percent of the time, and one of those hypotheses, a new mechanism for an E. coli protein, was later backed up by an independent lab working the same problem.

In genomics the model spent more than a week working largely on its own, assembled single-cell data across 138 animal species, and trained a custom model that beat a system recently published in the journal Science despite being 100 times smaller.

The Safety Split

This is where the two-model design earns its keep. Anthropic says a Mythos-class model is capable enough to give real uplift to bad actors in cybersecurity and in biology, so Fable 5 ships with a layer of classifiers that watch for risky requests. When a query trips the filter on cybersecurity, on biology and chemistry, or on attempts to distill the model into a competitor, Fable 5 hands the response off to Opus 4.8 instead, and the user is told every time that handoff happens.

The company tuned these filters on the cautious side, which means some harmless questions get caught, though it reports that more than 95 percent of sessions never hit a fallback at all and that the filters trigger in under 5 percent of sessions on average.

An external bug bounty running more than 1,000 hours turned up no universal jailbreak. Anthropic also set a new rule that all traffic on Mythos-class models will be held for 30 days to help catch novel attacks, with that data walled off from training and deleted afterward.

What It Costs You

Both models run at 10 dollars per million input tokens and 50 dollars per million output tokens, which lands at under half the price of the earlier Mythos Preview. For readers on a subscription, there is a window worth marking.

Fable 5 is included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans from launch through June 22, and on June 23 it shifts to paid usage credits before Anthropic aims to fold it back into standard plans once it has the capacity to handle the demand.

The Bottom Line

Fable 5 is the first time Anthropic has put frontier, above-Opus capability in front of everyone, and the science results are the part that should make you sit up rather than the coding speed. The honest tension is the rerouting, since the model you pay for will sometimes quietly answer through a weaker one, and how often that bites will decide whether the safety split reads as responsible or as annoying. Mark June 22 on your calendar either way.

Category: Technology Strategy

"I'm deciding whether my team at [Company] should adopt a new frontier AI model for [specific workflow, e.g. code migration]. Walk me through a decision framework that weighs capability gains against cost per million tokens, the risk of requests being rerouted to a weaker model, and a possible pricing change after [date]. End with three questions I should answer before committing."

The interesting line today was not a benchmark. It was a model writing a biology hypothesis that a separate lab then confirmed without knowing where it came from. We have spent two years watching these systems get better at the work we already do, and this is one of the first times one of them produced an idea worth testing on its own. Worth sitting with as the rerouting debate plays out. Hit reply, I read every response.

See you in the next one.

— Vivek

P.S. Know a founder or engineer who keeps a close eye on model releases? Forward this their way. They can subscribe at https://savvymonk.beehiiv.com/