Press Space to continue
Model routing and cost optimization create a critical tradeoff
Press Space to continue
On one hand this makes sense... most folks will just pick the biggest SOTA model when for a given task that's not needed On another, companies can secretly up charge and you would never know that you got routed to a crappy model and got charged SOTA prices Not saying Lovable would do this fyi
Anton Osika – eu/acc@antonosikaSep 1
The model picker era is coming to an end. More than 1B+ prompts have been sent to Lovable, and each one has helped us learn which model performs best for each task. For every prompt in Lovable, our system decides the right models, tools, instructions, and context - when to retry, when to replan, and when to route work to our own trained models. A model company can improve its model, but we improve the whole system, combining all the frontier models. In the end, we don’t want you to think about models. We just want you to go build cool stuff. Read in full: https://lovable.dev/blog/the-model-picker-is-a-dead-end
