Grok API Pricing Calculator

How Grok API pricing works in 2026

xAI bills Grok by the token, not by the chat. You pay for input (the prompt), cached input (prompt tokens served from cache), and output (the completion). Rates are listed per million tokens on the official models page. Grok 4.6 shipped August 12, 2026 as the flagship for coding and long-running agents. A request that also fires server-side tools — web search, X search, code execution, file search — pays a second bill on top of tokens. Generic calculators that only multiply tokens by a list price miss both the 200K long-context cliff and those tool fees.

Grok model comparison: what each actually costs

Below 200K prompt tokens: Grok 4.6 is $2 / $0.50 / $6 (input / cached / output per 1M). Grok 4.5 is the same $2 / $6 with a cheaper $0.30 cache. Grok 4.3 and every dated Grok 4.20 SKU (reasoning, non-reasoning, multi-agent) are $1.25 / $0.20 / $2.50 with a 1M context window. Grok Build 0.1 is $1 / $0.20 / $2 at 256K context. At or above 200K, every one of those rates doubles. OpenRouter lists the same Grok 4.6 pass-through rates ($2 / $0.50 / $6). Grok 4.1 Fast is not on the current official pricing table, so it is omitted here.

The 200K trap

This is not an overflow surcharge. Official docs say requests whose prompt reaches 200K tokens are billed at the higher rate for all tokens in the request. Crossing the line by a few tokens can double input, cache, and output. If your agent stuffs a repo into the prompt, stay under the line or budget 2×.

Cached input

Prompt caching is the only token column where Grok 4.6 is dearer than 4.5 ($0.50 vs $0.30 per 1M). Repeat the same system prompt and you pay the cache rate on the hit portion. A 50% hit rate on an 8K prompt is real money at 100 requests a day. Cache rates also double if the prompt crosses 200K.

Tool calls

Web Search, X Search, and code execution are $5 per 1,000 successful calls. File attachment search is $10 per 1,000. The model decides how many times to call; an agentic research loop can add more than the token bill. Collections / RAG search is $2.50 per 1,000 and is not in this calculator. Image and video understanding on search results are token-billed, not per-call.

Worked example

The default shape on this page is 8,000 input tokens, 2,000 output tokens, 100 requests a day, 30 days, Grok 4.6, no cache, no tools, under 200K. That is $0.028 per request and $84.00 this month. Flip the 200K toggle and the same month becomes $168.00. Add one web search per request and tools add $15.00. Switch the same shape to Grok Build 0.1 and the token bill drops to $36.00.

Grok vs GPT vs Claude

Frontier token prices move. This page does not invent competitor rates. For a side-by-side of Grok against GPT and Claude on the same workload, use aipricingcalculators.com.

SuperGrok vs API break-even

SuperGrok is $30 per month on the official Grok plans page — a consumer chat plan with a weekly usage pool, not an API key. This calculator’s default Grok 4.6 workload is $84 on the API, so a SuperGrok seat is cheaper if you only need the consumer product. Break-even on the API is $30 divided by cost per request (about 1,071 requests per month at the default $0.028). They are different products. Do not treat SuperGrok as a substitute for production API traffic. SuperGrok Heavy is listed at $300 per month.