Anthropic has introduced new flagship models Claude Fable 5 and Claude Mythos 5, setting a new benchmark for capabilities and cost in the frontier model segment.
What Happened
Anthropic introduced Claude Fable 5, available to a wide range of users, and Claude Mythos 5, which is open only to Project Glasswing participants and features less stringent safety restrictions. Both models support a 1 million token context window and are equipped with the Adaptive Thinking feature. Usage costs have increased significantly: $10 per 1 million input tokens and $50 per 1 million output tokens, which is twice the cost of the current top-tier Opus 4.8.
Context
As part of its new strategy, Anthropic is building a clear hierarchy of models: Sonnet is focused on efficiency, Opus on the premium segment, while Fable and Mythos form a new segment of ultra-powerful and expensive AI agents for solving the most complex tasks.
Why It Matters for the Industry
For the industry, this means the creation of a new class of high-performance 'reasoning engines'. The appearance of the stop_reason: "refusal" parameter in the API upon a successful HTTP 200 response requires developers to immediately revise their response handling logic and agent architectures.
Why It Matters for Users
Users gain access to next-level tools for deep research and autonomous programming. However, using these models will require careful prompt optimization and cost-benefit analysis due to the sharp increase in token costs.
What Is Not Yet Known / Limitations
There is a significant cost barrier for small teams and solo builders, which may limit the widespread adoption of these models in smaller projects.
Sources
Author
Look at AI, Editorial Team