Weave Router Routes LLM Prompts to Right Models, Cutting Costs 40-70%

Weave has launched the Router, a lightweight open-source model orchestration tool designed to route AI prompts to the most suitable large language model in under 50 milliseconds. Developed by Weave, a prominent engineering intelligence platform, the router addresses increasing enterprise demands for cost optimization and multi-model flexibility in agentic workflows. Rather than relying on heuristic or prompt-based selection, the system utilizes a compact local embedder paired with a performance-optimized cluster scorer to dynamically assign requests to the optimal model based on real-time capability matching. The platform operates as a drop-in proxy compatible with major cloud providers, including Anthropic, OpenAI, and Gemini, while also supporting open-weight models like Llama, Mistral, and Qwen through OpenRouter or standard OpenAI-compatible endpoints. It natively handles streaming, function calling, and multimodal inputs, and maintains strict data governance by keeping provider API keys encrypted directly on user infrastructure. Built-in observability automatically exports OpenTelemetry traces to third-party monitoring stacks or Weave’s native dashboard, enabling granular tracking of latency, token usage, and cost attribution. Independent benchmarks position the Router at the top of the RouterArena leaderboard, achieving a score of 76.09 in the Acc-Cost Arena category. Engineering teams at organizations such as Robinhood, PostHog, and Reducto have reported infrastructure cost reductions between 40 and 70 percent without compromising response quality. The tool supports rapid deployment through an npx-based installer or a self-hosted configuration, featuring automated patching for developer workflows including Claude Code, OpenAI Codex, opencode, and Cursor. By abstracting model selection into a standardized API endpoint, the Router enables development teams to dynamically switch providers, test alternative architectures, and scale inference workloads with minimal code modifications. The release establishes a new baseline for standardized, cost-efficient AI routing protocols as organizations increasingly deploy polyglot model ecosystems in production environments.