Google Reports 2.85× Faster Gemma 4 E2B Inference With Gemini 4 Argon

Google says Gemini 4 Argon helped optimize Gemma 4 E2B inference to run 2.85 times faster on an NVIDIA A100 using vLLM, lifting reported throughput from about 37,900 to more than 108,000 tokens per second. The company says the change preserved token parity and output quality.