Grok 4.3 Beta Opens to Subscribers as Musk Says 1T Flagship Nears Training Completion

xAI is pushing Grok beyond conversational AI with Grok Bot, an autonomous agent designed to operate across applications and keep working around the clock. Each bot receives a dedicated cloud computer, allowing it to carry out multi-step tasks more like a digital employee than a chatbot. The launch underscores a broader industry shift toward agents that can execute work continuously, raising the competitive stakes for products from Anthropic, OpenAI and other AI developers.

Grok Bot entered public testing in August, while Elon Musk said Grok 4.6 would arrive later the same week. Subsequent reports said xAI had released the model, with benchmark results matching GPT-5.6 in some tests and stronger coding capabilities becoming available through Cursor. Investor Gavin Baker called the product xAI’s “Claude Code moment” and said it had increased his own AI usage by roughly 100 times, though the reported benchmark and productivity claims remain vendor- or user-reported.

Grok is a central part of Elon Musk’s push to compete in the fast-moving large language model market. The contest with Kimi K3 matters beyond headline model size, as developers increasingly weigh reasoning performance against computing costs and token efficiency. Grok 4.6 is described as a 2-trillion-parameter model trained with SpaceX engineering data, while Kimi K3 has 2.8 trillion parameters.

Musk said on X that Grok 4.6 would be released two weeks after his announcement, with Grok 4.7 following four weeks later, though he did not provide exact calendar dates. The stated goal is for Grok 4.6 to match or surpass Kimi K3 in reasoning while maintaining high token efficiency. No training cost, pricing details or comprehensive benchmark results were disclosed with the launch timetable.