Anthropic Launches Claude Fable 5, An AI That Plays Pokémon And Actually Codes
An AI startup just dropped a model that actually does what tech bros have been promising for years. It's time to retire the 'AI is just a fancy autocomplete' meme, because this new beast is already doing real engineering work while we sleep.
The payment giant Stripe utilized the model to migrate a massive Ruby codebase of 50 million lines in just 24 hours. Under normal circumstances, a team of human engineers would have spent over two months staring at screens, drinking cold coffee, and questioning their life choices to achieve the same result.
In another test, the model solved complex programming tasks with high efficiency. Instead of hogging massive computing power like a crypto mine in winter, it completed the tasks while keeping resource usage remarkably low.
For the gamers and visual researchers, the model proved its vision capabilities by playing Pokémon FireRed from start to finish. Unlike previous attempts that required a complex web of hand-holding instructions and external tools, this system navigated the game purely by looking at raw screenshots, proving that AI might soon navigate messy legacy software interfaces better than humans.
Behind the scenes, the company also developed a restricted version called Claude Mythos 5 for cybersecurity and biology researchers. In a joint program with the US government, this locked-down variant spent a week autonomously analyzing genomic data across 138 animal species and built its own machine learning model that outperformed a recently published study in Science magazine, despite being 100 times smaller.
In medical research, the restricted model acted as an autonomous bio-informatician, designing promising protein structures for neurodegenerative and muscular diseases. It even formulated a biology hypothesis about E. coli bacteria that was later independently confirmed by a completely different physical laboratory.
To prevent bad actors from using these chemical and biological superpowers for evil, the public version uses advanced classifiers that instantly spot dangerous requests. If someone asks for something sketchy, the system silently redirects the prompt to Claude Opus 4.8, a safety handoff that currently triggers in less than five percent of user sessions.
Both models are now priced at ten dollars per million input tokens and fifty dollars per million output tokens.
While the tech industry has spent the last year drowning in useless hype, this release marks a terrifyingly quiet transition from chatty assistants to autonomous scientific agents. It seems the future didn't arrive with a loud bang, but with a virtual monkey playing Game Boy and designing molecular cures on the side.
Source: Anthropic
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.