Weave Router is an open-source model router for agentic coding systems. It scores every prompt for complexity in milliseconds and routes each turn to the cheapest model that can handle it, escalating back to a frontier model when a task stalls, which its makers say cuts coding-agent costs by 40 to 70 percent with little more than an endpoint change. The project has more than 5,500 stars on GitHub.
Frontier models are wasted on easy turns, but naive downgrading breaks tasks and prompt caches. Weave Router addresses this with a complexity classifier trained on large volumes of real coding sessions, cache-aware decisions that only route down when the saving beats the cost of rebuilding the cache, and an escalation classifier that watches for stalls and loops. It can also drain the flat-rate subscriptions you already own, such as Claude and Codex seats, before spending metered API budget, and one command configures Claude Code, Codex and Cursor to use it. The project publishes its benchmark methodology openly, comparing routed runs against frontier-only baselines on terminal and codebase question-answering suites, so teams can judge the quality cost of routing for themselves rather than trusting the headline savings.
Freemium. The router itself is open source under the Apache-2.0 license; Weave also sells hosted coding plans, Code Max at $10 per month and Code Boost at $100 per month, plus enterprise deployment.
Developers and teams who spend heavily on coding agents and want to cut model costs without giving up frontier-model quality on the turns that matter.