Augment Prism routes models, cuts costs

This is a practical infrastructure play disguised as a product feature: if your agent sessions are long and expensive, smarter routing matters more than squeezing a few more points out of a single model.

• –Prism is cache-aware, which matters because switching models mid-session can blow away prompt cache savings and erase the routing win
• –Augment’s internal benchmark framing is credible for the use case, since it models multi-turn coding work better than one-shot evals
• –The tradeoff is obvious: routing only helps if the planner reliably detects when a cheaper model is “good enough” and avoids churn
• –The feature reinforces Augment’s positioning as a cost-optimized coding platform, not just another AI editor with model choice
• –For teams, the bigger implication is operational: model selection is becoming something the platform should abstract, not something every user should micromanage