Short answer
Grok 3 is no longer served as Grok 3 through the xAI API. Effective May 15, 2026, requests using the grok-3 model slug are automatically routed to Grok 4.3 with no reasoning effort. So for current budgeting, use the price of the model that serves the request—not an old Grok 3 rate. (docs.x.ai)
xAI lists Grok 4.3 API prices in USD per 1 million tokens:
| Prompt length | Input | Cached input | Output |
|---|---|---|---|
| Below 200,000 prompt tokens | $1.25 | $0.20 | $2.50 |
| 200,000 prompt tokens or more | $2.50 | $0.40 | $5.00 |
When a request reaches the long-context threshold, the long-context rates apply to all tokens in that request. (docs.x.ai)
What happens if your code still says grok-3?
xAI’s retirement notice says the grok-3 slug continues to resolve, but requests are served by grok-4.3 with none reasoning effort. That means a successful request to the old slug does not mean you are still using the Grok 3 model. For predictable production behavior, update your request to name grok-4.3 explicitly and budget at its rates. (docs.x.ai)
The official retirement notice identifies the redirect and replacement, but does not establish a current Grok 3 price. Search snippets showing other figures should not be treated as verified current xAI rates without a source that confirms the model, provider, and pricing period.
Estimate your API spend
For a request below the long-context threshold, estimate token charges as:
(uncached input tokens × input rate + cached input tokens × cached-input rate + output tokens × output rate) ÷ 1,000,000
Use mutually exclusive input counts in the formula. If your usage report gives total input tokens including cache hits, calculate uncached input tokens = total input tokens − cached input tokens. Choose the applicable rate tier using the entire prompt length, including cached input tokens—not just uncached input tokens.
For example, assume one request uses 100,000 uncached input tokens and 10,000 output tokens, with no additional billable tools:
- Input: 100,000 × $1.25 ÷ 1,000,000 = $0.125
- Output: 10,000 × $2.50 ÷ 1,000,000 = $0.025
- Estimated total: $0.15
This is an illustrative calculation using Grok 4.3’s listed short-context rates, not a Grok 3 quote. Actual usage depends on the tokens counted for the request and whether input is billed at the cached rate. xAI notes that API request processing can add predefined prompt tokens, so your billed token count may differ from a plain-text estimate. (docs.x.ai)
For production budgeting, estimate representative prompts and outputs, check the total prompt length, including cached input, against the 200,000-token threshold, and review actual usage in your API team’s billing records. If requests continue using the retired slug, price them as Grok 4.3 requests under the redirect behavior.
Is API usage included in a Grok subscription?
No. xAI’s API accounts FAQ says Grok and xAI API billing are separate. A shared account login does not combine the bills: API billing is managed in the xAI Console. (docs.x.ai)