Grok 4.7 pricing: $2 / $6, and the SKU people quietly swap
A price-comparison review, not a rate card. Grok 4.7 stays at $2 / $6 per million tokens, the same as Grok 4.6. The fast tier doubles that price. The ‘half the price’ line only holds against the Max SKUs in xAI’s own table.
| Bill | Grok 4.7 | The number people quote instead |
|---|---|---|
| Standard list, $ per 1M in / out | $2 / $6 | Same as Grok 4.6, not a discount |
| Fast variant, as we read the launch note | $4 / $12 | Twice the output speed, twice the price — confirm live |
| Claude Sonnet 5 list on this desk | $2 / $10 | Not the column in xAI’s table |
| Claude Opus 5 list on this desk | $5 / $25 | Not the column in xAI’s table |
| Fable 5.1 Max, xAI’s table | $10 / $50 | The Claude SKU the ‘half price’ line is aimed at |
| GPT-5.6 Sol Max, xAI’s table | $4 / $20 | The other named comparable |
| GitHub Copilot | Provider list price | Gradual rollout, not a cheaper SKU |
The number, dated
As of the 21 September 2026 launch note, Grok 4.7 starts at $2 per million input tokens and $6 per million output tokens. xAI says that is the same price and the same speed as Grok 4.6. You are not being offered a discount for upgrading. You are being offered a new model id at the old list price.
There is a second SKU. xAI serves a fast variant with twice the output speed at twice the price. Applied to the $2 / $6 list, that reading is $4 input and $12 output. The launch note does not print “$4 / $12” as a separate row. Confirm the live rate card before you put it in a budget. Source: xAI’s Grok 4.7 note.
What “half the price” is allowed to mean
The page headline says Grok 4.7 is twice as fast at half the price of comparable models. The table underneath names the comparables. GPT-5.6 Sol Max is $4 / $20. Fable 5.1 Max is $10 / $50. Against those two output prices, $6 is less than half of $20 and far below $50. Against Grok 4.6, the price is identical, so “half” cannot mean “half of last month’s Grok bill.”
It also cannot mean half of Claude Sonnet 5. This desk’s Claude profile dates Sonnet 5 at $2 / $10 and Opus 5 at $5 / $25. Input matches Sonnet. Output is cheaper than Sonnet’s $10 ($6 versus $10) and cheaper than Opus. That is a modest output gap, not a half-price story. Teams that paste the Fable Max column into a Sonnet migration memo are mixing SKUs. The comparison page is there to keep those columns apart.
A flat price is not a flat invoice
- Output dominates agent runs. At $2 / $6, a long coding session is an output bill. Self-checks, retries, and “works longer on hard tasks” — xAI’s own description of 4.7 — push that side of the invoice. We have not metered 4.7’s tokens per task. Do not let a blog do it for you.
- The fast tier is a different product. Twice the speed at twice the price only helps if the run was latency-bound. If you were token-bound, you just doubled the bill.
- Cache, batch, and reasoning tokens are not in the headline. Other labs’ cheap-looking list prices move the same way. If xAI publishes cache or batch rates later, this page should be redated. It does not invent them today.
- Copilot is not a second, cheaper Grok. GitHub says Copilot bills Grok 4.7 at provider list price under usage-based billing, and the rollout is gradual. Seats do not include a private discount that this desk can see. GitHub Changelog, 21 Sep 2026.
When the cheap list price is the wrong reason
Switching to Grok 4.7 because Terminal-Bench or CursorBench must go up is not supported by xAI’s own Fable column. Those rows favor Fable 5.1 Max. Switching because EEBench or the Harvey legal-agent row must go up is the argument the sheet actually supports, and it is still a vendor exhibit. Switching because the API price undercuts a $50 output SKU is arithmetic. Switching because a headline said “half the price of comparable models,” without naming the comparable, is how invoices surprise people.
The model profile, including what 4.7 is not, is here. Jev’s price review stays a different question: typed decisions versus generation.