Dari Launches Open-Source AI Router Scoring 79.8% on Coding Benchmark for $76
Dari launched a cache-aware router model today that scores 79.8% on the Terminal-Bench 2.1 coding benchmark for $76 in total inference costs. The small, fine-tuned model dynamically assigns coding steps to cheaper language models and only calls expensive frontier models when the route actually changes the output.
To avoid the cost spikes caused by resetting prompt caches mid-session, the router only switches models when a cheaper path saves money. The weights are on Hugging Face, and the system plugs into Claude Code, Codex, and Pi, allowing developers to bring their own Anthropic and OpenAI subscriptions. On the same 89-task suite, direct frontier model runs cost into the thousands.
From the sources (2 posts)
@kimmonismus79.8% on Terminal-Bench 2.1 for $76 in total inference cost. That's the number Dari @daridotdev is launching with today. They open-sourced the router model behind it: a small fine-tuned model that decides which model handles each step, sen
@omarsar0Switching models mid-session is one of the fastest ways to inflate an agent bill. Every switch resets your prompt cache, and a cheaper model can end up costing you more. @daridotdev built their router to be cache-aware. It only switches