Gemini 4 Argon is coming out, but access to it will remain paid.

Gemini 4 Argon has begun rolling out in the Gemini app and Antigravity, but only paying users will be able to try the model. Even on lower-tier plans, access will be limited: the cautious launch quickly gave way to expansion, but Argon still never became free.

Google has not announced any changes to the model's capabilities. The company continues to cite previously published benchmarks and, to demonstrate its usefulness, provides examples from its own development work.

According to Google, Argon helped reduce the load on YouTube's servers by 45 percent after detecting redundant emoji rendering. In the physics simulator for Project Wing drones, the model found critical bugs that had been sitting in the code since 2018. These are compelling cases for internal tasks, but they do not show how the model will handle everyday requests from ordinary users.

Google also discusses safety measures: the model is trained to refuse malicious instructions, and an automated monitor can stop processing if it detects an undesirable workflow. The article contains no independent assessments of these mechanisms. Access is paid, lower-tier plans are limited, and claims about effectiveness and safety rely on Google's own data.

If you pay for AI tools, how important are these internal Google cases to you when choosing a model? Source: AI - Ars Technica.