Grok API Pricing: Per-Model Cost Per Token

Batch requests bill 20% cheaper, and cached input (tokens you resend, like a fixed system prompt) bills at a quarter of standard. The headline rate is where an estimate starts, not where the bill lands.

We ran three Grok models on the same job and measured what each run actually cost. Bubble size is the per-token price, so Grok 4.6 and 4.7 are the same size. Yet 4.7 sits far to the right, because it spends more tokens to answer and costs about five times what 4.3 did for the same work. The per-token price never shows you that. The cost of a real run does.

We scored those same runs against a correct answer. The first job was an everyday one, pulling action items out of a meeting note; the second had a single right answer, working out ship dates under a set of rules. On the everyday job the three models landed within five points of each other. On the harder one they all tied at 100. Grok 4.3, the cheapest in the lineup, cleared the bar both times, so paying up for a pricier Grok bought no more quality here.

xAI updates these models constantly, so the model that clears your bar today can slip below it next week while a rival climbs past it. Measuring once picks the right model now. Measuring on a schedule keeps quality high and cost down with no quiet regressions or cost creep.