xAI is selling more capability at Grok 4.6's headline price, while its own long-context schedule shows why base token rates are an incomplete measure of enterprise cost.
xAI launched Grok 4.7 on September 21 at $2 per million input tokens and $6 per million output tokens, the same standard tariff it publishes for Grok 4.6. A workload containing 1 million input and 1 million output tokens costs $8 at the base rate before caching, tool charges or long-context pricing. For enterprise buyers, context length is already an operating cost variable before model quality enters the comparison.
That corrects the most consequential ambiguity in xAI's launch pitch. The company describes Grok 4.7 as offering roughly twice the speed at half the price of comparable models, but its own release says the standard model is served at the same price and speed as Grok 4.6 while producing better benchmark results. Relative to its predecessor, the commercial offer is more capability for the same base tariff.
The pricing table becomes less simple once prompts grow. xAI's release notes list Grok 4.7 at $4 per million input tokens and $12 per million output tokens for prompts above 200,000 tokens, with cached input at $1. Those rates are twice the standard input and output prices. Long-context agent workloads can cross into a materially different cost tier even when the headline tariff is unchanged.
The model war is moving from token price to completed-task cost
A stronger model can still lower the cost of work at the same token rate if it needs fewer retries, shorter repair loops or less human intervention. The reverse can happen when reasoning traces, long context and repeated tool calls expand faster than task quality improves. Two models with identical headline token prices can produce very different costs per accepted coding task, research job or support resolution.
xAI makes latency another explicit trade. Grok 4.7 Fast is offered at roughly twice the output speed and twice the standard token rates. It is available only through Cursor and Grok Build and is not offered on the public xAI API. Lower latency is therefore a separately priced service available through two distribution surfaces rather than a blanket speed upgrade for API customers.
Amazon's Bedrock distribution currently applies to Grok 4.6. AWS added that model on August 19, placing xAI inside the same procurement, identity, logging and governance layer enterprises use to compare other model providers. That lowers the operational friction of testing xAI and also makes model substitution easier when several vendors can perform the same task.
Grok 4.7 gives enterprise buyers two distinct price signals. The standard tariff did not fall from 4.6, while capability rose; the long-context schedule can double published rates once prompts pass 200,000 tokens. The useful enterprise benchmark is the cost and latency of an accepted task under the workload's actual context and retry pattern.
Sources
https://x.ai/news/grok-4-7 https://docs.x.ai/developers/grok-4-7 https://docs.x.ai/developers/release-notes https://aws.amazon.com/about-aws/whats-new/2026/08/amazon-bedrock-grok-4-6/