Google Releases Gemini 4 Argon for Coding and Security Work
- Google DeepMind released Gemini 4 Argon, which ties for first on CWE-Bench at 68% for vulnerability remediation and is described as its most resilient model against indirect prompt injections.
- In Chrome, Argon helped the team finish an architectural refactor across hundreds of files in a ~6 million line legacy folder, completing months of work in weeks.
- Using Argon, the team reached a 2.85x end-to-end inference speedup on an A100 via vLLM, raising Gemma 4 E2B throughput from ~37,900 to over 108,000 tokens per second with 100% token parity.
- Argon's xTB changes made the quantum chemistry package up to 2.3x faster at predicting molecular structures and 1.8x faster at conformer search, and the changes were contributed to the official GitHub repository.
- Access is rolling out to Ultra and Pro subscribers, but Hacker News commenters reported Pro users in the EU and US who could not see Argon in the Gemini app or Antigravity.
Hacker News opinions
Same here, our org is in the EU and Argon doesn't show up at all.
Pro user here and I can't see it in the Gemini app or the Antigravity CLI. AI Studio keeps pushing me to upgrade to Ultra.
OpenRouter had it live pretty fast, so the API side seems ahead of the consumer apps.
Argon showed up in Antigravity for some Pro users after an update, so it looks more like messy rollout than a hard lock.
The product page is painful on mobile. My phone ran out of memory and forced a refresh every time I scrolled.
I care most about the writing. It's clear and not RL-fried, and the em dash is back, which honestly made me laugh.
Coding looks close to solved, but the gap between releases keeps shrinking. I'd take 10x tok/s over another point bump.
The old Pro habit of numbering answers with labels like The Parameters seems to have eased off a bit in this one.