Grok 4 Fast is not a dependable free production API option anymore. xAI retired the grok-4-fast and grok-4-1-fast API models on May 15, 2026. Requests using those old model names now route to Grok 4.3, which has its own usage-based charges. The old model name still resolving does not mean the old model or its pricing is still in effect. [xAI’s retirement notice]
Free access to Grok through the consumer service is a separate offer: xAI lists a $0/month plan with limited usage. That does not mean API calls are free, nor does it confirm that the retired Grok 4 Fast model is available in the consumer app. [xAI pricing]
Free Grok access and free API usage are different
| What you want to use | What “free” means | Budget takeaway |
|---|---|---|
| Grok consumer app | xAI lists a free plan with limited usage. The offer is for the consumer service; it does not promise access to a particular retired model. | Fine for trying Grok within the plan’s limits; don’t treat it as free API capacity. [xAI pricing] |
| xAI API | API usage is billed per token. xAI’s quickstart says to load credits before making API requests. | Budget for API charges; don’t assume a free credit or quota. [xAI API quickstart] |
| Platform free-start offers | A platform may advertise a free start, but that message alone doesn’t say whether model inference is included, what limits apply, or who pays. MindStudio’s Grok 4 Fast listing advertises starting for free but says its pricing is not published in the available metadata. | Check the platform’s billing terms; don’t treat a free-start promotion as a confirmed free usage allowance. [MindStudio model listing] |
| User-pays intermediary | Puter describes its Puter.js arrangement as one where developers can use the integration without managing API keys or paying usage themselves, while each user covers usage through their Puter account. |
The tutorial does not specify end-user rates or a free allowance, and does not establish zero-cost Grok 4 Fast inference. This is Puter’s arrangement—not direct xAI API billing. [Puter tutorial] |
Bottom line: Consumer free access, a platform promotion, an intermediary’s user-pays arrangement, and direct API billing are different channels. “Grok is free to start” does not establish that production API calls are free.
What does the Grok 4 Fast API cost now?
The API model IDs grok-4-fast-reasoning, grok-4-fast-non-reasoning, grok-4-1-fast-reasoning, and grok-4-1-fast-non-reasoning were retired on May 15, 2026. xAI says requests to those slugs are routed to grok-4.3 and billed at Grok 4.3 rates, not at the retired models’ former rates. [xAI’s retirement notice]
The published Grok 4.3 rates are:
| Token type | Price per 1 million tokens (USD) |
|---|---|
| Input | $1.25 |
| Cached input | $0.20 |
| Output | $2.50 |
For requests whose prompt reaches the 200,000-token long-context threshold, the listed rates are $2.50 per million input tokens, $0.40 per million cached input tokens, and $5.00 per million output tokens. The long-context pricing rule applies the higher rates to all tokens in a request once its prompt reaches the threshold. xAI also lists a 1.1× rate for requests to its US regional endpoint. [xAI API pricing]
Example: estimate a small API workload
Suppose one request uses 100,000 uncached input tokens and returns 20,000 output tokens. It stays below the 200,000-token long-context threshold. Using the published Grok 4.3 rates:
- Input: 100,000 ÷ 1,000,000 × $1.25 = $0.125
- Output: 20,000 ÷ 1,000,000 × $2.50 = $0.05
- Estimated model-token total: $0.175
This is an illustrative calculation, not a promise of a fixed bill. It covers model-token charges only; it does not include any separately applicable tool or platform charges. Cached input, long-context requests, and endpoint choice can also change the estimate. [xAI API pricing]
What to check before deploying
- Use the model ID you actually intend to pay for. Don’t leave a retired Grok 4 Fast slug in production assuming it will keep its old behavior or price. xAI says those slugs now route to Grok 4.3. [xAI’s retirement notice]
- Estimate with current token categories and thresholds. Count input and output separately, account for cached input where applicable, and check whether the prompt reaches the 200,000-token long-context threshold. [xAI API pricing]
- Treat consumer plans, platforms, and the API as separate budgets. A limited free consumer plan or a platform’s “get started free” message does not guarantee free API inference. xAI’s API quickstart instructs developers to load credits before using the API. [xAI pricing; xAI API quickstart]
- Check endpoint choice. xAI lists a 1.1× rate for requests to its US regional endpoint. [xAI API pricing]
Bottom line
You can start using Grok through xAI’s limited free consumer plan, but that is not the same as free production API usage. Grok 4 Fast’s API models have been retired; old API slugs redirect to Grok 4.3, billed at $1.25 per million input tokens and $2.50 per million output tokens at standard short-context rates, with higher rates for long-context requests. Budget against the current model and pricing rules—not an old “Grok 4 Fast free” claim. [xAI’s retirement notice; xAI API pricing]