Weave Router

Weave Router is a routing layer for coding agents, built by Weave Engineering Intelligence. It sits between a coding agent and the model providers, reading each turn the agent takes and choosing which model should answer it. The product is aimed at engineering teams that already hold subscriptions with several model vendors and want to use them together.

What Weave Router does

• Per-turn routing: scores how hard each step is with a classifier, sends easier steps to cheaper models and keeps the frontier models for the turns that need them.
• Subscription-aware routing: sends work to whichever model plan a team holds that still has quota left, including models from one vendor used inside another vendor's agent.
• Cache-aware switching: moves a conversation to a different model only when the saving is larger than the cost of rebuilding the prompt cache.
• Vendor comparison: the company has published a run against GPT-6 Astra on the same Codex harness, which it describes as its own measurement.

Who it serves

Engineering teams that run coding agents and pay for more than one model subscription. The company describes coding harness support on its product pages.

Key facts

• Version: 2.0, released 16 September 2026.
• Company: Weave Engineering Intelligence, whose main product measures engineering work.
• Benchmarks: the company's own figures on Terminal-Bench 4.0 and SWE-Atlas, from its own harness.