tencent/TCAndon-Router
TCAndon-Router is a 4 billion parameter reasoning-centric multi-intent routing module developed by Tencent, designed for agent selection in multi-agent systems. It supports dynamic agent onboarding without retraining and provides transparent, interpretable routing decisions. This model excels at resolving agent conflicts and achieves state-of-the-art performance on real-world enterprise datasets, making it ideal for complex intent-routing scenarios.
Loading preview...
What the fuck is this model about?
TCAndon-Router is a 4 billion parameter model developed by Tencent, specifically designed as a reasoning-centric multi-intent routing module. Its primary function is to select appropriate agents in multi-agent systems, but it can also be applied to any intent-routing scenario, including agent skill selection. The model is trained using Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (DAPO).
What makes THIS different from all the other models?
Unlike many general-purpose LLMs, TCAndon-Router is purpose-built for agent routing in multi-agent systems and enterprise applications. Its key differentiators include:
- Dynamic Agent Onboarding: New agents (intents) can be added simply by appending their descriptions, without requiring retraining of the model.
- Interpretable Routing: Provides transparent routing decisions, enhancing explainability, robustness, and cross-domain generalization.
- Conflict Resolution: Effectively handles agent conflicts by preserving all relevant agents when multiple are applicable, allowing a downstream Refining Agent to merge outputs.
- State-of-the-Art Performance: Achieves top results on large-scale, real-world enterprise datasets like HWU64, MINDS14, SGD, and Tencent Cloud ITSM (QCloud), often outperforming models like GPT-5.1 and Claude-Sonnet-4.5 in these specific routing tasks.
Should I use this for my use case?
Yes, if your use case involves:
- Multi-agent systems: You need to intelligently route user queries to the most appropriate agent(s).
- Intent routing: You have various intents or skills and need a robust system to classify and direct user input.
- Enterprise applications: You require a solution designed for real-world enterprise scenarios with dynamic changes and a need for explainability.
- Dynamic environments: Your system frequently adds or modifies agents/intents and you want to avoid constant retraining.
This model is particularly strong where precise, interpretable, and adaptable intent classification is critical for orchestrating complex AI workflows.