Direct answer
There is no current xAI API rate for serving the original Grok 4 model. The legacy model ID grok-4-0709 was retired on May 15, 2026. Requests that still use that ID are automatically routed to Grok 4.3 and billed at Grok 4.3 rates—not at the original Grok 4 model’s rates. xAI’s migration documentation lists those redirected-request rates as $1.25 per million input tokens and $2.50 per million output tokens. (docs.x.ai)
So if you’re budgeting a production integration, treat a grok-4-0709 request as a legacy alias for Grok 4.3, not as access to the original Grok 4 model. The current xAI pricing page lists Grok 4.3, Grok 4.5, Grok 4.6, Grok 4.7, and other models—but no current Grok 4 entry. All listed prices are in USD. (docs.x.ai)
What the current charge covers
For a request routed to Grok 4.3 on the standard global API route, the official price table lists these token rates: (docs.x.ai)
| Billed category | Price per 1 million tokens, short context |
|---|---|
| Input | $1.25 |
| Cached input | $0.20 |
| Output | $2.50 |
When a prompt reaches the model’s 200,000-token long-context threshold, xAI applies the long-context rates to all tokens in that request: $2.50 per million input tokens, $0.40 per million cached input tokens, and $5.00 per million output tokens. The published context window for Grok 4.3 is 1 million tokens. (docs.x.ai)
The US regional endpoint carries a 10% premium over the global token rates. If you use it, include that multiplier in your estimate. Server-side tools can also add separate invocation charges on top of token usage, so include them when they’re enabled for your workload. (docs.x.ai)
Estimate a workload
For requests below the long-context threshold, a basic estimate is:
(uncached input tokens ÷ 1,000,000 × $1.25) + (cached input tokens ÷ 1,000,000 × $0.20) + (output tokens ÷ 1,000,000 × $2.50)
Example: Suppose a service handles 1,000 requests in a month. Each request uses 10,000 input tokens and returns 1,000 output tokens; assume no cached input, no tool calls, and prompts below 200,000 tokens.
- Input: 1,000 × 10,000 = 10 million tokens → 10 × $1.25 = $12.50
- Output: 1,000 × 1,000 = 1 million tokens → 1 × $2.50 = $2.50
- Estimated monthly model-token spend: $15.00
That example uses Grok 4.3 replacement rates, not an original Grok 4 rate. It excludes separately billed tools and any regional premium. For prompts at or above 200,000 tokens, use the long-context rates for every token in that request rather than applying short-context rates to the first 200,000 tokens. (docs.x.ai)
Which model and API route should you budget for?
- Still sending
grok-4-0709? The retired slug continues to resolve, but xAI says requests are redirected togrok-4.3with low reasoning effort and billed at Grok 4.3 prices. For predictable production behavior, update the model field to the model you intend to use instead of relying on the legacy redirect. (docs.x.ai) - Want current xAI API pricing? Budget using the exact model ID and route in your application. Grok 4.3’s rates are not Grok 4.5 or Grok 4.7 rates; model versions have separate pricing. (
docs.x.ai) - Using a third-party API provider? Treat that provider’s price and model ID as a separate offer. For example, Venice documents
grok-4-5, which is Grok 4.5 on Venice—not xAI’s legacy Grok 4 API. Do not substitute its rates for xAI’s or for Grok 4. (docs.x.ai)
A subscription price is also not the same thing as an API token rate. For software sending API-key requests, estimate against the applicable API model and provider pricing—not a consumer plan fee.
Bottom line: xAI’s documented treatment of legacy grok-4-0709 requests is Grok 4.3 service at $1.25/M input and $2.50/M output tokens on the standard short-context global route. Add cached-input pricing, long-context rates, regional premiums, and tool charges where they apply; don’t label the replacement rates as the price of the original Grok 4 model. (docs.x.ai)