Grok 4.7 API pricing and cost calculator

How this is calculated

A request with 100,000 input and 5,000 output tokens costs $0.20 for input and $0.03 for output when nothing is cached, which makes $0.23 per call. Reading half the input from cache brings the call to $0.155, and 100 calls a month come to $15.50. Past 200,000 input tokens the whole request moves to $4 input, $1 cache reads and $12 output per million, not just the tokens beyond the line. A 300,000 input and 5,000 output request crosses it, so it costs $1.26 with no cache hits and $0.81 with half the input cached. Agent work is where the cache rate earns its place. A coding session of 60 turns that each send 50,000 input tokens, mostly cached, and write 1,200 output tokens costs $2.382 in total. The same session with no cache hits costs $6.432. Batch jobs have their own discount. Both input and output come in 20% below standard. A nightly review job of 5,000 requests with 30,000 input and 3,000 output tokens each costs $312 in batch and $390 at standard rates.

Verdict

Price Grok 4.7 on finished tasks. A coding or agent task usually takes many calls, and each call adds its own input, cache reads and output. Record those totals on a sample of your real work, then multiply by your volume. Keep the cached prefix stable, since a prefix that changes every turn can't be read from cache. Watch where your longest prompts land against the 200,000 token line, because crossing it reprices the whole call. Ask xAI's documentation whether cached tokens count toward that line before you design around it. Image and file inputs turn into tokens too, so measure them on real samples. If you're deciding whether to move up from Grok 4.6, run the same prompts through both and compare cost per accepted result, not cost per call.