Google announced the global availability of its Gemini 4 Argon model on October 13, following an initial limited release to cybersecurity partners in late September. The company positions Argon as a "frontier model" designed for complex, multi-step tasks in domains such as coding, finance, legal, cybersecurity, and science. Google DeepMind Senior Vice President Koray Kavukcuoglu stated that Argon excels at deep work and demonstrates precision across these workflows.
According to Google, Gemini 4 Argon outperforms rival models like OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 and Claude Opus 5.5 on various internal benchmarks. The model reportedly ranks first on the Vals Index, which assesses performance across economic sectors including finance, coding, legal, and tax work. Google also claims Argon leads on Zapier's AutomationBench, a benchmark for end-to-end execution across business functions, with a score of 51.3%. In coding, Google states Argon achieved a 77.9% score on DeepSWE v1.1, a benchmark for long-horizon software engineering tasks. The company further noted a 35% increase in agentic code submissions and a 30% decrease in post-submission code rollbacks by its internal engineering teams using Argon over a four-week period.
Independent evaluations offer a more nuanced perspective on Argon's performance compared to its competitors. Artificial Analysis, for instance, assigned Argon an Intelligence Index score of 53, placing it roughly on par with GPT-6 Astra but behind Claude Opus 5.5 (58) and Claude Sonnet 5.5 (56). However, Artificial Analysis also noted Argon's lowest hallucination rate among high-scoring models, at 15%.
Google has highlighted Argon's capabilities in cybersecurity defense, stating the model can autonomously find, validate, and patch critical software vulnerabilities. The company reported that Argon successfully solved all 12 simulated cyberattack scenarios in internal tests, an increase from five in the previous generation. Additionally, Google states Argon is its most resilient model against indirect prompt injection attacks, with a reported success rate of 0.7% for such attacks on Gray Swan's benchmark.
Gemini 4 Argon is now available to developers and enterprises through the API in AI Studio and the Gemini platform. Google AI Ultra subscribers can access the model in Google Antigravity, while Google AI Ultra and Pro subscribers can use it in the Gemini app and AI Mode in Search. Google has set an introductory API price of $2 per million input tokens and $10 per million output tokens, with cached input tokens at a 95% discount. This introductory pricing is scheduled to expire on November 30, 2026, after which standard rates will double to $4 per million input tokens and $20 per million output tokens. Google DeepMind Vice President Tulsee Doshi referred to Argon as a "powerhouse" model and indicated it is the first in a new series of Gemini models.