While Google is preparing for the public launch of the new AI model Gemini 4, which is being developed under the code name Argon, company employees are testing more advanced versions of the neural network. According to Business Insider, a more advanced version of Gemini 4, known as Carbon, is available to Google employees, which should help the company close the gap with competitors in the field of AI.
Image source: Igor Omilaev / unsplash.com
How asserts source, a new version of Gemini 4 a few days ago became available to Google employees on the internal Jetski platform, which the company uses to generate program code. According to one Google employee, Carbon’s AI model “feels like Opus 5.5” in programming tasks. We are talking about Anthropic’s most productive model for agent programming. However, he added that the AI model needs to be tested for a longer period of time.
Google’s internal testing shows the company is making rapid progress toward creating the next generation of Gemini 4 AI models, which it announced late last month. It also underscores how actively Google is in the race to become the leader in AI programming agents. Anthropic and OpenAI are currently leaders in this space with their models designed for complex engineering work.
Early versions of Argon tested internally were generally well received by employees. At the same time, some of them noted that early versions of the AI model lagged behind competitors’ advanced neural networks in some tasks of writing program code. More advanced versions of Gemini 4 could close the gap, but there’s no guarantee Google will release them. According to the source, the company tested several versions of Gemini 4, which were developed under the code names Argon, Barium and Carbon.
Image source: BoliviaInteligente/unsplash.com
Google’s internal codenames are usually not the same as public codenames. For example, one of the documents indicates that the Barium-B model was chosen as the one that would become publicly known as Argon. It is currently unknown whether Carbon will be another update to the neural network under the name Argon or whether it is a separate AI model in the Gemini family.
Last month, Google said its Argon AI model performs on par with cutting-edge neural networks in benchmarks for programming and professional tasks, including legal and finance. The company noted that at Argon, the developers placed special emphasis on cybersecurity. At the initial stage, the model will be available to Google partners as part of the Fairwind program, an initiative to search for vulnerabilities in new models. Later the model will be available to the general public.
Google has been under increasing pressure lately to show that it can still compete with the leading players in the AI field. This week, Google announced a new Gemini-based general purpose AI agent for running workloads. In general, the race for leadership in the segment of agent-based AI assistants between the largest companies is becoming increasingly heated.
If you notice an error, select it with the mouse and press CTRL+ENTER.
