Weave Router Optimizes LLM Costs for Coding Agents

Weave has introduced a new model router designed to optimize the use of large language models (LLMs) within coding agents like Claude Code, Codex, and Cursor. This router acts as an intermediary, analyzing each inference request and dynamically selecting the best LLM to handle it, prioritizing both cost-efficiency and performance. The system employs a reinforcement learning model, trained on extensive agent traces, to make intelligent routing decisions. For instance, complex planning tasks might be sent to a powerful model like Opus 4.8, while simpler code exploration or implementation steps could be routed to faster and cheaper alternatives. Weave reports internal savings of 40% on token costs without compromising quality or development velocity. The router is available as a source-available tool under the Elastic License 2.0 for self-hosting, or via a hosted service.