- Updated: November 25, 2025
- 2 min read
Salesforce AI Research Unveils xRouter: A Cost‑Aware RL‑Based LLM Orchestration Engine
Salesforce AI Research Unveils xRouter: A Cost‑Aware RL‑Based LLM Orchestration Engine
Salesforce AI Research has introduced xRouter, a groundbreaking reinforcement‑learning router designed to orchestrate large language models (LLMs) while keeping inference costs under control. Built on the open‑source Qwen2.5‑7B‑Instruct model, xRouter can intelligently select from a pool of more than 20 LLMs, deciding when to answer a query directly and when to delegate it to a more suitable model.
The router continuously learns from usage patterns, optimizing both performance and expense. In internal benchmarks, xRouter achieved near‑GPT‑5 level accuracy while reducing the overall inference cost by up to 80%. This makes it a powerful tool for enterprises that need high‑quality AI responses without the prohibitive price tag of running the largest models for every request.
Key capabilities of xRouter include:
- Cost‑aware routing: Dynamically balances response quality against compute cost.
- Reinforcement learning: Continuously improves routing decisions based on real‑world feedback.
- Model diversity: Supports a catalog of 20+ LLMs, allowing fine‑grained selection for different tasks.
- Self‑answering: Determines when it can answer directly without invoking an external model.
For a deeper dive into the technical details, read the original announcement on MarkTechPost.
Explore related insights on our site:
Stay tuned for more updates on how xRouter can transform AI‑driven applications by delivering smarter, cheaper, and faster language model services.