The routing methodology works across various model combinations, shifting the performance-cost frontier for tasks involving code, chat, and work. For instance, pairing Fable 5 and Kimi K3 through Max resulted in a 10.7% net improvement and a 54% cost decrease compared to Fable 5 High, with costs dropping to $1.06 per task. Arena is currently providing direct access to the Fable 5.1 and Astra variant in Agent Mode and plans to release an API in the future.
Tuesday, October 20, 2026
Story
Arena's Max Router Matches Claude Fable 5.1 Performance at 37% Lower Cost
AI7 days ago
An agentic model router called Max now allows users to automatically select the most efficient AI model for a given task based on output quality and price. Developed by Arena and trained on 2.5M real-world agentic sessions, the tool recorded a 14.7% net improvement score in its internal benchmark. When routing tasks between Claude Fable 5.1 and GPT 6 Astra, Max matched the output of Fable 5.1 at a median cost of $3.27 per task, representing a 37% reduction in price.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.