Google's internal AI testing shows progress in coding models

Google employees are testing advanced versions of its Gemini 4 model, with internal codenames like Carbon and Barium, as the company races to improve its AI capabilities. According to Business Insider, Carbon, possibly an update to Argon, was made available on Google's internal platform, Jetski, and is being compared to Anthropic's Opus 5.5 for coding tasks. One employee noted that Carbon 'feels like Opus 5.5' but emphasized the need for further testing. Google's internal testing highlights its rapid development of Gemini 4 models, which were announced in late September. The company is under pressure to compete with rivals like Anthropic and OpenAI, which have made strides in AI coding agents. Google's latest updates may close performance gaps, though the release of these models remains uncertain. The company also highlighted Argon's focus on defensive cybersecurity and its planned rollout to partners before a public launch. Google recently announced a new 'universal' Gemini AI agent for the workplace, as competition in agentic assistants intensifies. Internal discussions among staff suggest optimism about the latest updates, with one employee calling Carbon 'really good.' The company's efforts to enhance its AI models continue as it aims to maintain its position in the fast-evolving AI landscape.