The AI race just hit a new gear. Anthropic has officially unveiled its next-generation "Mythos-class" models: Claude Fable 5 and Claude Mythos 5. This isn't just another incremental update; it's a fundamental shift from AI that chats to AI that acts.
If you've been following the evolution of Large Language Models (LLMs), the term "agentic AI" has been the holy grail. With today's launch, that future seems to have arrived.
Beyond the Chatbox: What is Mythos-Class?
For the past year, Anthropic's Opus was the gold standard for reasoning. However, the new Fable 5 and Mythos 5 represent a tier of capability that Anthropic calls "Mythos-class." These models don't just process information; they can work autonomously for longer periods, use their own notes to improve outputs, and handle "long-horizon" tasks that used to require human teams.
Claude Fable 5: The New King of Benchmarks
Fable 5 is the model now available to the public, and its stats are staggering. It currently leads nearly all tested benchmarks for AI capability, excelling in areas where previous models struggled:
- Software Engineering: Stripe reported that Fable 5 compressed months of work into days. In one test, it migrated a 50-million-line Ruby codebase in just 24 hours—a task that would typically take a human team months.
- Vision-Based Mastery: While earlier models needed specialized "harnesses" to navigate complex visual tasks, Fable 5 successfully played Pokémon FireRed from start to finish using nothing but raw screenshots. No maps, no navigation aids—just pure visual reasoning.
Claude Mythos 5: The Lab Partner of the Future
While Fable 5 is for everyone, Claude Mythos 5 is a specialized version currently being deployed through Project Glasswing in collaboration with the US government. It has the safeguards lifted in key areas to assist cyberdefenders and scientific researchers.
The results in the life sciences are already borderline science fiction. Mythos 5 isn't just looking up data; it's proposing novel scientific hypotheses. In one genomics study, the model assembled data for millions of cells across 138 species and trained its own custom machine learning model that outperformed existing models published in journals like Science.
Safety with a "Fallback" Twist
Releasing a model this powerful (especially one that can autonomously find software vulnerabilities) requires extreme caution. Anthropic has implemented a clever safety mechanism: The Opus 4.8 Fallback.
Instead of just refusing a query that triggers a safety classifier (in areas like cybersecurity or biology), Fable 5 automatically passes the request to Claude Opus 4.8. This ensures that users still get a helpful, safe response rather than a frustrating "I can't help with that" message. Anthropic notes that this fallback triggers in less than 5% of sessions.
How to Get Claude Fable 5
If you're looking to upgrade your workflow, here’s the breakdown of availability:
- Claude API: Developers can start using
claude-fable-5 today.
- Pricing: $10 per million input tokens and $50 per million output tokens—slashed significantly from previous preview pricing.
- Subscription Plans: Pro, Max, and Team users have free access to Fable 5 until June 22, 2026. After that, it will transition to a credit-based system until capacity scales up further.
Anthropic's latest move puts a massive amount of power into the hands of developers and researchers. Whether you're migrating a massive legacy codebase or designing the next generation of therapeutics, Fable 5 is officially the new state-of-the-art.
Stay tuned for our full technical breakdown of Fable 5's coding efficiency and the Pokémon vision challenge.
