BetaRouting models for cost, speed, and quality
Send heavy reasoning to frontier models and fan lightweight subtasks to faster providers — without changing your prompt.
By The MasterNode Team · Guides
One model for everything is simple and expensive. Parallel workflows are the opposite: different subtasks have different needs. MasterNode lets you bring keys for multiple providers and route work accordingly.
Match the model to the subtask
- Planning and final merge: use your strongest reasoning model.
- Bulk extraction, formatting, and transforms: Groq or DeepSeek for throughput.
- Embeddings and retrieval: Cohere or your provider's embedding endpoint.
- Vision over slides or screenshots: a multimodal model on the branches that need it.
Parallelism changes the math
Four cheap parallel calls can finish before one huge frontier call and still cost less. Watch usage per tenant in Settings and adjust routes after you see which stages dominate latency and token count.
No lock-in
Swap providers when pricing or quality shifts. Your prompts, agents, and memory stay put — only the inference backend changes.
Try MasterNode
Bring your own model keys and run your first parallel workflow in minutes.