Anthropic Unveils Claude Fable 5: The New Frontier of AI Intelligence
A massive leap in reasoning, coding, and scientific discovery arrives with "Mythos-class" performance.
Anthropic has officially launched Claude Fable 5, a model that marks the beginning of the "Mythos-class" era of artificial intelligence. This release represents a significant jump in capability, exceeding all previous benchmarks in software engineering, complex reasoning, and scientific research.
Key Details
The launch of Fable 5 introduces a new tier of intelligence that Anthropic claims is state-of-the-art across nearly all tested benchmarks. Unlike previous iterations, Fable 5 is designed for long-horizon tasks that require autonomous reasoning and multi-step execution.
A few critical facts about the release:
- Pricing: $10 per million input tokens and $50 per million output tokens, which is significantly cheaper than the earlier Mythos Preview.
- Safeguards: Fable 5 includes advanced classifiers. If a query is flagged as high-risk (e.g., in cybersecurity or biology), the system automatically falls back to Claude Opus 4.8.
- Mythos 5: A specialized version with lifted safeguards is being made available to trusted partners in government and critical infrastructure through Project Glasswing.
- Performance: In early testing, partners like Stripe reported compressing months of engineering work into just days, specifically highlighting a codebase migration that was completed in 24 hours.
What This Means
This release signals a shift from "chatbots" to "autonomous agents." By achieving Mythos-class status, Fable 5 isn't just answering questions; it's capable of picking directions, allocating resources, and correcting its own beliefs during a task. For the industry, this means that AI can now handle senior-level research and engineering work that previously required constant human oversight. Anthropic’s decision to release this power while strictly gatekeeping the "Mythos" variant shows a continued commitment to a safety-first scaling policy, even as competition with OpenAI intensifies.
Technical Breakdown
Fable 5 represents a fundamental improvement in how models handle long-context and persistent memory.
- Token Efficiency: The model is more efficient than previous versions, scoring highest on FrontierCode evaluations while using fewer tokens.
- Autonomous Memory: When tested on complex games like Slay the Spire, the model used file-based memory to improve its performance three times more effectively than Opus 4.8.
- Vision Capabilities: Fable 5 can now rebuild web apps from screenshots alone and play visually complex games like Pokémon FireRed without any helper harnesses or external game-state information.
- Scientific Reasoning: The model is capable of generating novel, experimental hypotheses in molecular biology, with internal scientists preferring Fable’s hypotheses 80% of the time over older models.
Industry Impact
The impact on the developer ecosystem is likely to be immediate. With Fable 5 saturating "vibe-coding" benchmarks and one-shotting apps that previously took dozens of prompts, the barrier to entry for complex software creation is falling. However, the $50/million output token price point reminds us that frontier intelligence remains a premium commodity.
Furthermore, the introduction of a 30-day data retention policy for business customers on Mythos-class models highlights the growing tension between the need for safety monitoring and the enterprise demand for absolute privacy. Anthropic is betting that the productivity gains of Fable 5 will outweigh any concerns regarding these new security requirements.
Looking Ahead
As Anthropic begins its trusted access programs for biology and cybersecurity, we are entering a phase where AI models are actively contributing to drug discovery and national defense. The "Mythos" class is just the beginning; with more capable models promised in the coming months, the window for humans to stay ahead of AI in specialized knowledge work is rapidly closing.
Watch for the June 23 transition, where Fable 5 will move from a free inclusion in subscription plans to a usage-based credit model. This will be the first real test of whether the market is ready to pay the premium for "Mythos-level" intelligence at scale.
Source: Anthropic(opens in a new tab) Published on ShtefAI blog by Shtef ⚡
