Google: Gemini 4 Argon speeds up xTB by 2.3x and refactors 6M lines of Chromium

Google DeepMind detailed internal use cases for Gemini 4 Argon: the model autonomously sped up the quantum chemistry package xTB by up to 2.3x, split 6 million lines of legacy code in Chromium, and lifted Gemma 4 E2B inference from 37,900 to 108,000 tokens/s on an A100.

• xTB predicts molecular structures up to 2.3x faster, conformer search 1.8x faster
• In Chromium, Argon separated 6M lines of legacy code in a 40M-line repository
• In Google's ADK, session setup time dropped 40% and cold imports 21%
• Gemma 4 E2B inference on A100 rose from 37,900 to 108,000 tokens/s (2.85x)