Google revealed its most powerful model – and what's the connection to Israel?

Google announced Gemini 4 Argon, its first flagship model of the fourth generation, capable of generating a continuous output of one million tokens. The development is accompanied by a collaboration with the Israeli cybersecurity company Wiz, which used it to identify critical vulnerabilities in hospitals. The model is currently available only to a limited group of partners under the Fairwind program.

Google announced Gemini 4 Argon, the first flagship model in the Gemini 4 family, developed at Google DeepMind. The model boasts an unprecedented output capability of one million tokens in a single sequence – a 15-fold leap over previous generations limited to 64,000 tokens. In performance benchmarks, Argon surpassed OpenAI's GPT-6 Astra (77.9% vs. 74.1% on the DeepSWE v1.1 test) and Anthropic's Claude Opus 5.5 (74.2%). However, it does not lead in all areas: on the Terminal-Bench 4.0 test, Anthropic's model leads (66.4% vs. 57.4%). The model is linked to a notable Israeli collaboration: the cybersecurity company Wiz harnessed it under the Scan for Good initiative to scan critical public infrastructure, revealing a severe security flaw in medical software used by hospitals worldwide. In blind penetration tests, the model achieved a 70.9% success rate in attacking live websites. Currently, the model is available only to a limited group of cybersecurity partners under the Fairwind program, subject to voluntary review by US security and government agencies. When opened for broad use, the price will be $2 per million input tokens and $10 per million output tokens, with a 95% discount for cache memory reads.