01 What happened
Google has released additional performance and safety details for its Gemini 4 Argon model, including an 87% score on the Embodied Reasoning (ERQA) benchmark and a 0.7% attack success rate on the Gray Swan Indirect Prompt Injection test. The company claims safety violations have been reduced nearly 2-fold compared to previous Gemini models. Separately, Google is offering 10,000 scientists six months of free access to the highest usage tier of Google Antigravity.
02 Key details
- Google reports Gemini 4 Argon scored 87% on the Embodied Reasoning (ERQA) benchmark, outperforming other frontier models on pure reasoning in 3D environments.
- On the Gray Swan Indirect Prompt Injection benchmark, Argon achieved a 0.7% attack success rate across 15 adaptive attempts, leading all evaluated models.
- Safety evaluations showed external bio red teaming unlocked less than 1% of tested bio sub-tasks, and adversarial stress-test violations were reduced nearly 2-fold versus previous Gemini models.
- Google is offering 10,000 scientists six months of free access to the highest usage tier of Google Antigravity, subject to eligibility, capacity, and selection; applications are open.
03 Why it matters
The new benchmarks and safety evaluations provide concrete evidence of Argon's capabilities in physical reasoning and security, crucial for enterprises and researchers considering deployment. The free scientist program aims to accelerate research.
04 Who it matters to
Enterprises, software developers, researchers and information security specialists.