Google Launches Gemini 4 Argon, Most Advanced AI Model
Alphabet has unveiled Gemini 4 Argon, its first flagship AI model in seven months, designed to rival OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5. Argon excels in cybersecurity, scoring 68% on CWE-bench v1 and 77.9% on DeepSWE-v1.1, and matches GPT-6.1 Sol on a key security test. Independent benchmarks score it 53 points, trailing Claude Opus 5.5 but matching GPT-6 Astra. The model is rolled out cautiously, initially available to cybersecurity partners and for U.S. government safety testing, with no set date for full enterprise release. Google uses Argon internally to optimize data center memory. Pricing is $2 per million input and $10 per million output tokens (50% launch discount), expected to double later, with release scheduled after October 1, 2026. Meanwhile, Google's Spark lags behind Meta's Muse (5 million downloads) and OpenAI's Dots in the personal agent market. Google has cancelled Gemini 3.5 Pro amid DeepMind restructuring and researcher exodus.
- •Alphabet unveiled Gemini 4 Argon, its first flagship AI model in seven months, initially available to cybersecurity partners and for U.S. government safety testing, with no set date for full enterprise release.
- •Argon excels in cybersecurity (68% on CWE-bench v1, 77.9% on DeepSWE-v1.1) and matches OpenAI's GPT-6.1 Sol on a key test, while scoring 53 points on independent benchmarks (matching GPT-6 Astra but trailing Claude Opus 5.5).
- •Pricing is $2 per million input and $10 per million output tokens (50% launch discount), with prices doubling later; release is scheduled after October 1, 2026.
