Google announces Gemini 4 Argon model, competes with GPT-6 Astra

Google recently officially announced its highly anticipated new artificial intelligence model, Gemini 4 Argon. This model, which has a great impact in the technology world, leaves its competitors behind with its performance especially in complex workflows, cyber defense and software development. Gemini 4 Argon, which is currently undergoing testing by the US government and whose wide-scale access is still limited, competes directly with OpenAI's GPT-6 Astra model with the results it achieved in artificial intelligence benchmark tests. This performance increase, which confirms Google's internal RSI (self-improvement of the artificial intelligence system) rumors, signals the beginning of a new era in artificial intelligence technologies.

According to data shared by Artificial Analysis, Gemini 4 Argon made a big leap, especially in agentic evaluations. Offering significant improvements in areas where previous Gemini versions were weak, the model outperformed its competitors by scoring 77.5% on AutomationBench-AA. The system, which has an assertive position in challenging tests such as Terminal Bench 4, also draws a balanced profile in Analytical Quality and Presentation Quality criteria.

Gemini 4 Argon is the first non-Flash class model offered by Google after a seven-month break. The model, which reached the same level as the GPT-6 Astra with 53 points in the artificial intelligence index, also follows a very competitive pricing strategy in terms of cost. The model, initially offered with an entry fee of $2 and an exit fee of $10 per 1 million tokens, aims to be among the market averages even after the discount period ends.

Although the test results are impressive, there are some question marks regarding the real-world performance of the model. A report published by Bloomberg stated that some employees within Google expressed concerns that the model was having difficulty performing certain coding tasks. This shows that academic achievements may not always be fully reflected in practical applications.