Elon Musk's AI company has released Grok 4.7, its newest model for coding and general knowledge work. It costs $2 for every million words of text it reads and $6 for every million words it writes, a price that undercuts nearly every major American competitor.
That pricing is not new. xAI has held the same rate across several model generations now, and it looks like a deliberate strategy to stay the cheapest option among the big American AI labs, even if it still costs more than the fully open Chinese models it is trying to outprice. A Chinese model called DeepSeek V4.1 Flash charges a small fraction of what xAI charges per word, and it still edged past Grok 4.7 on a coding test.
The bigger issue is what you actually get for that price. On a combined test covering ten different reasoning and knowledge tasks, Grok 4.7 landed well behind OpenAI's GPT-6 and Anthropic's Claude Fable 5.1. The gap widens sharply on real coding work done inside a computer terminal, where Grok 4.7 finished less than half the tasks that its rivals completed.
There is a reason for that weakness worth understanding. Musk had promised Grok 4.7 would ship around September 12, but the launch slipped. According to Musk's own public posts, the delay came from a training mistake: the process that was supposed to make the model reason more carefully instead punished it for writing long answers, so the model learned to give up early on hard problems and skip double-checking its own work. That is exactly the opposite of what xAI advertised the model would do, and it is exactly the kind of behavior a business does not want from software handling tasks without a human checking every step.
This launch also cannot be separated from who owns xAI now. Earlier this year, SpaceX absorbed xAI in an all-stock deal, making the AI unit a wholly owned part of Musk's rocket company. SpaceX trades publicly, and Wall Street is already weighing Grok's performance in its outlook on the stock, with at least one major bank staying bullish on the idea that cheap, capable AI tools could widen SpaceX's business well beyond rockets and satellites. In other words, rocket and satellite profits now partly fund a price war in AI subscriptions, a subsidy most pure AI companies do not have.
Grok 4.7 also launched right after both OpenAI and Anthropic released stronger models within roughly the same two weeks, resetting expectations for the whole market almost overnight. That pace is the real story here. Whichever AI model your business picked six weeks ago may already be behind, and today's best model will likely be passed again before your next budget cycle closes.
For any business comparing AI vendors, price is the easiest number to compare and often the worst way to decide. If the work involves writing code, running multi-step research, or handling tasks without constant human review, a model that finishes a quarter of test tasks instead of most of them will cost more in cleanup and correction than it ever saved in subscription fees.