Anthropic cuts Fable 5 costs with Sonnet delegation
Curated by the Inblix editorial team
Anthropic has a smart workaround for Fable 5’s steep price tag: use it like a boss who only steps in for the tough decisions. Instead of letting Fable 5 handle everything, the company now recommends two patterns where smaller, cheaper models like Sonnet 5 do the heavy lifting. In the ‘Advisor’ setup, Sonnet runs tasks and only pings Fable when it’s stuck, hitting 92% of Fable’s solo SWE-bench Pro score for 63% less cost. The ‘Planner’ pattern flips it—Fable maps out the plan, then Sonnet agents execute, achieving 96% of BrowseComp performance at nearly half the price. Both patterns use separate caches to save on context costs. Anthropic’s likely motivation? Mounting price pressure from Chinese open-source models and the newly released GPT-5.6 Sol, which is cheaper per token and more efficient. Why it matters: This reveals a growing trend in AI—making frontier models viable by turning them into specialists rather than jack-of-all-trades workhorses.
💡 Key Takeaways
- Anthropic now recommends using Fable 5 as a planner or advisor, delegating execution to the cheaper Sonnet 5 model.
- The Advisor pattern achieves 92% of Fable 5's SWE-bench Pro performance at 63% of the cost.
- The Planner pattern delivers 96% of BrowseComp performance at 46% cost savings by using Sonnet 5 for execution.
- Anthropic's cost-saving advice is driven by competition from cheaper Chinese open-source models and the more efficient GPT-5.6 Sol.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.