Browser LLM Evaluation

Cloud (OpenRouter)

Model openai/gpt-4o-mini meta-llama/llama-3.2-1b-instruct google/gemma-3n-e2b-it:free meta-llama/llama-3.2-1b-instruct meta-llama/llama-3.2-3b-instruct API Key OpenRouter

On-Device

Model (transformers.js) gemma-3-270m-it-ONNX gemma-3-1b-it-ONNX Llama-3.2-1B-Instruct-ONNX Llama-3.2-3B-Instruct-ONNX gemma-3n-E2B-it-ONNX (not working) Load Model

Request Pattern

1 request / sec Every 10 sec 1 request (not implemented) Batch: 10 every 5s (not implemented) Burst: 50 then idle Route strategy Always device Always cloud Round Robin Probabilistic (p to cloud) Cloud probability (for probabilistic)