Anthropic just dropped two new AI models, and the pitch is simple: these things don’t lose focus on Tuesday when you started the project on Saturday. Claude Fable 5 and Claude Mythos 5, unveiled on June 9, are designed for developers and advanced users who need sustained performance across complex, multi-day tasks.

The pricing tells a story on its own. At $10 per million input tokens and $50 per million output tokens, both models come in at less than half the cost of the previous Mythos Preview. In a market where AI companies typically charge more for better models, Anthropic is doing the opposite, cutting prices while claiming superior long-horizon capabilities.

What Fable 5 actually does differently

Early developer feedback points to strong performance specifically in long-horizon reasoning and complex coding tasks. The model exceeds 90% on Anthropic’s core benchmark for complex tasks, a threshold that its predecessor, Opus 4.8, apparently struggled to match consistently over extended interactions.

Fable 5 is the general availability version, meaning it ships with Anthropic’s standard safety measures baked in. It’s accessible through the Claude API, Claude.ai, AWS Bedrock, Google Cloud, Microsoft Foundry, and various partner marketplaces.

Mythos 5 is being offered to select partners without the same guardrails that Fable 5 carries. That dual structure, a safe public model paired with a more flexible partner-only tier, is a deliberate strategic choice that lets Anthropic maintain its safety-first reputation while still competing on raw capability.

The competitive landscape is getting crowded

By making Fable 5 cheaper than its own predecessor, Anthropic is trying to lower the barrier for power users who might otherwise hesitate to run long, token-heavy sessions. The previous Mythos Preview sat at more than double these rates. Developers who were rationing their API calls or batching queries to keep costs manageable now have significantly more room to let the model work extended problems without watching the meter spin.

The broad availability across AWS Bedrock, Google Cloud, and Microsoft Foundry means an enterprise team on Azure, AWS, or Google Cloud can try Fable 5 without rearchitecting anything, reducing the switching cost to near zero.