Amazon Cuts Costs Swapping Claude Sonnet for Strands Decider
Original: Amazon saves millions by replacing Sonnet 4.6 with Strands Decider
Why This Matters
Model routing is emerging as a primary cost-control lever for enterprises running agents at scale.
Amazon replaced Claude Sonnet 3.5 v2 (Sonnet 4.6 internally) with Strands Decider, a lightweight routing model, saving millions in inference costs by reducing unnecessary calls to expensive frontier models.
Amazon has deployed Strands Decider, a purpose-built routing component within the Strands agent framework, to replace direct reliance on Anthropic's Claude Sonnet as the default decision-making model in its internal agent pipelines. The move reportedly saves Amazon millions of dollars by steering only complex tasks to expensive frontier models while routing simpler decisions to cheaper alternatives. Strands Decider functions as an intelligent dispatcher: it evaluates an incoming task and selects the appropriate model tier rather than defaulting every call to a high-cost LLM. The blog post, published on strandsagents.com, frames the savings as a direct result of this tiered model-routing architecture. Strands Agents is Amazon's open-source agent framework, and the Decider component sits inside its harness layer. The case study underscores a broader industry push toward model routing and cost arbitrage—using smaller, faster models for classification and planning steps, reserving frontier calls for tasks that genuinely need them. No specific dollar figure beyond 'millions' was disclosed, and the timeframe for the savings was not stated.