Cactus Unveils Hybrid AI Router

Cactus Hybrid Router, announced by Cactus Compute, is an ultra-small model with 65k parameters that determines the execution path in real time between local edge models (Gemma4-2B, etc.) and cloud frontier models (Gemini-3.1-Flash-Lite, etc.) depending on the difficulty of the task.

Key features are as follows:

• Performance and Cost Optimization: By routing 15-55% of tasks to the cloud, it achieves Gemini-3.1-Flash-Lite-level performance while primarily using local models.
• Quantization Resilience: Through its proprietary Cactus Quants technology, it maintains near-FP16 performance even under 4bit quantization.
• Flexible Resource Allocation: The edge-cloud routing ratio can be adjusted to optimize the balance between infrastructure cost and response speed.
• Multimodal Support: It can process not only text but also Vision and Audio prompts.

This technology aims to reduce the inference burden on cloud infrastructure and accelerate the expansion of AI to edge devices.