task-model-router

You can route multi-model requests by inspecting message content for image blocks and code signals, applying token thresholds for long context, and selecting the best-fit model for each specific task type automatically.

Task classification for model routing detects whether a request involves vision, code reasoning, large-document summarization, or general chat to select the optimal LLM. You need it when using a single model for every request causes quality loss or inefficiency.

To handle mixed inputs like images and code blocks, the system applies deterministic routing rules to detect specific content triggers, routing each part to the most suitable model while optimizing both speed and cost.

If the selected model fails during multi-model routing, the system applies fallback routing to automatically redirect the request to the next best available model, ensuring task completion.

Yes, you can use model routing for long documents by applying token and length thresholding to detect when inputs exceed small-model context limits, then routing them to a model that supports large context.

The best way to optimize LLM costs for mixed content is to use deterministic task classification, routing simple chat to fast models and reserving expensive vision or reasoning models only for inputs that require them.