Weave Router

Weave Router is a model router for coding agents from Weave Engineering Intelligence, the Y Combinator-backed company at weaveos.com whose main product measures engineering work. Version 2.0 was released on 16 September 2026. It sits between a coding agent and the model providers: it reads each turn the agent takes, scores how hard the task is with a classifier, and routes the request to the cheapest model that can handle it, keeping the frontier models for the turns that need them. It is subscription-aware, so a team can use one vendor's models inside another vendor's agent on the plans it already holds, routed to whichever still has quota left, and its cache-aware switching only moves a conversation to a different model when the saving beats the cost of rebuilding the prompt cache. Weave publishes a comparison run against GPT-6 Astra on the same Codex harness, the same 66 tasks and two attempts each, paired on the same day: Router solved 62.1% to Astra's 60.6%, at 20.5 minutes per trial against 44.1, and roughly half the cost per trial. Those are the company's own figures, from its own harness. The company also cites Terminal-Bench 4.0 and SWE-Atlas in its launch post. Weave's platform pricing is a free Starter tier, Pro at $50 per engineer a month and a custom Enterprise tier; the Router itself carries no published price and is sold by talking to the company. Not to be confused with the several unrelated products that share the name Weave.