Gemini 4 Argon with 1M-token context and advanced reasoning rolls out globally

Gemma 4 E2B: Inference through vLLM on an NVIDIA A100 achieved a 2.85x end-to-end speedup, raising throughput from around ...