Token prices per model, November 2026
All prices are in US dollars per million tokens, from xAI’s developer pricing page. The long-context rate applies once a request reaches 200k prompt tokens, and then to every token in that request.
| Model | Context | Input | Cached input | Output | Input / cached / output from 200k |
|---|---|---|---|---|---|
| grok-4.7 | 500k | $2.00 | $0.50 | $6.00 | $4.00 / $1.00 / $12.00 |
| grok-4.6 | 500k | $2.00 | $0.50 | $6.00 | $4.00 / $1.00 / $12.00 |
| grok-4.5 | 500k | $2.00 | $0.30 | $6.00 | $4.00 / $0.60 / $12.00 |
| grok-4.3 | 1M | $1.25 | $0.20 | $2.50 | $2.50 / $0.40 / $5.00 |
| grok-4.20-0309-reasoning and non-reasoning | 1M | $1.25 | $0.20 | $2.50 | $2.50 / $0.40 / $5.00 |
| grok-4.20-multi-agent-0309 | 1M | $1.25 | $0.20 | $2.50 | $2.50 / $0.40 / $5.00 |
| grok-build-0.1 | 256k | $1.00 | $0.20 | $2.00 | $2.00 / $0.40 / $4.00 |
Batch requests get 20 percent off on the grok-4.x models; results usually arrive within 24 hours and batch traffic does not count against your rate limits.
The costs that are not tokens
Agentic requests call server-side tools, and each call has its own fee on top of tokens.
| Tool or resource | Price |
|---|---|
| Web Search | $5 per 1,000 calls |
| Code Execution | $5 per 1,000 calls |
| File attachments search | $5 per 1,000 calls |
| Collections Search | $2.50 per 1,000 calls |
| X Search | $5 per 1,000 posts, $10 per 1,000 profiles |
| Remote MCP tools | No call fee, tokens only |
| File storage | $0.025 per GiB per day |
| Collections storage | $0.10 per GiB per day |
| Downloads | $0.20 per GiB |
A model that decides to search five times in one answer bills five tool calls. When you attach files to a request, Grok runs attachment_search as often as it thinks it needs to, and each run is a tool call.
A worked example
Hypothetical workload: an internal support agent on grok-4.7 handles 1,000 questions a day. Each question sends 20k tokens of instructions and context and gets 1k tokens back, and makes two Collections searches.
- Input: 20M tokens x $2.00 = $40.00
- Output: 1M tokens x $6.00 = $6.00
- Collections Search: 2,000 calls x $2.50 / 1,000 = $5.00
- Total: about $51 per day, before caching
If the 20k of instructions is identical each time and served from cache, most of that input drops to $0.50 per million, and the input line falls to roughly a quarter. Caching repeated context can reduce input costs; its effect depends on cache hits and the share of output and tool costs.
The common mistake: paying long-context rates by accident
Because the 200k threshold reprices the whole request, a team that stuffs every document into every prompt pays double on all tokens, not just the overflow. Keeping prompts under 200k, and fetching only the relevant pieces through Collections or retrieval, usually costs less than the “send it all” approach, even after paying for search calls.
Which model to start with
Compare grok-4.3, grok-4.20, grok-4.7 and grok-build-0.1 on your own tasks, using the table to account for token prices and context limits. grok-4.7 costs 1.6 times as much for input and 2.4 times as much for output as grok-4.3, so the quality gain has to be worth it for your work. A Dutch version with more on model choice is at wat kost de Grok API.
What this means for a team
Repeated instructions and documents add input costs; the largest cost category depends on the workload. Each agent your team builds on Grok carries its own copy of that context, and so does every agent built on Claude or OpenAI next to it. The Grok Enterprise pricing page covers what the chat seats add on top.
Where Bombos fits
Bombos gives every agent and AI tool a team uses one shared memory, so an agent can fetch the few pieces of team knowledge and learned rules a task needs instead of carrying the whole background in every prompt. It works with Grok, Claude, OpenAI models and your own agents, in our cloud or on your own server. A single developer can try it first with their own agents.
Related questions
Is the Grok API included in a Grok Business seat?
No. Grok Business and Enterprise seats cover chat in grok.com and the apps. API usage is billed separately per token through the xAI console.
What happens when a request reaches 200k prompt tokens?
Every token in that request, not just the part above 200k, is billed at the long-context rate. For grok-4.7 that is $4.00 input, $1.00 cached and $12.00 output per million.
Do Remote MCP tools cost extra?
There is no per-call fee for Remote MCP tools on the pricing page; you pay for the tokens the tool definitions and results add to the request.
Is there a cheaper model for coding?
grok-build-0.1 is listed at $1.00 input and $2.00 output per million tokens with a 256k context window. No batch discount is listed for it.
