Google unveils Gemini 4 Argon, direct challenge to OpenAI and Anthropic

Google has made public Gemini 4 Argon, the company’s first flagship model following Gemini 3. The launch comes as the large language model sector experiences a phase of intense rivalry among leading labs: OpenAI with GPT-6 Astra and Anthropic with Claude Opus 5.5 are identified as the direct competitors in the segment of the most advanced models.

On the technical side, Gemini 4 Argon achieves a score of 77.9% on the DeepSWE v1.1 test, one of the metrics used to evaluate software code comprehension and generation capabilities. On the CWE-bench benchmark, which instead measures the ability to identify vulnerabilities in code, the model ranks first, tied at 68%. Google has also extended the maximum output limit to one million tokens, a threshold that allows for generating responses and documents far longer than in previous versions.

On the commercial front, the launch price is set at $2 per million input tokens and $10 per million output tokens. This is a relevant figure for developers and companies integrating these models into their products, since the cost per processing unit is one of the factors guiding the choice between competing providers.

It should be noted that the news currently comes from a single source (AI Weekly); no independent confirmation is available. Google had not, at the time this article was written, released a separate official statement with the same technical details, and the benchmark results cited have not been verified by independent third parties.

The comparison between the three flagship models — Gemini 4 Argon, GPT-6 Astra and Claude Opus 5.5 — now hinges on specific performance in coding tasks and on the ratio between cost and processing capacity, two variables that client companies will use to decide which infrastructure to adopt in the coming months.