Short answer
Don’t start a new production integration on the original Grok 4 API model. xAI’s migration notice says the API slug grok-4-0709 was retired on May 15, 2026. Requests to it still resolve, but are redirected to Grok 4.3 with low reasoning effort and billed at Grok 4.3 rates. A successful response from the old slug therefore does not mean you are still running the original Grok 4 model. For a new integration, select a current model explicitly; if migrating, test the replacement against your workload. (docs.x.ai)
This guide concerns the direct xAI API, not consumer subscriptions or third-party platforms. The model family name “Grok 4” should not be confused with the retired API identifier grok-4-0709.
What happened to the Grok 4 model?
Grok 4 launched in July 2025, but that historical launch does not establish current API availability. The migration notice identifies grok-4-0709 as retired and says requests to it are routed to grok-4.3. For this retired reasoning-model slug, the redirect uses low reasoning effort. The notice says to select a model explicitly to control the replacement and reasoning effort rather than relying on the redirect. (docs.x.ai)
The redirect can preserve basic request continuity, but it does not guarantee the original model’s behavior, reasoning settings, or price. Update the model field and test the selected replacement before relying on it in production.
Which current model should you evaluate?
For a workload being migrated from grok-4-0709, start by testing Grok 4.3, the documented redirect target. Its model documentation lists text and image input, text output, a 1-million-token context window, function calling, structured outputs, and configurable reasoning effort. These are Grok 4.3 capabilities—not proof that the retired Grok 4 model had identical behavior. (docs.x.ai)
If choosing from scratch, consider Grok 4.7 as another candidate. xAI’s model documentation describes it as a flagship model and lists a 500,000-token context window. Those provider descriptions do not establish that it will perform best on your particular workload. (docs.x.ai)
| If your situation is… | Sensible next step |
|---|---|
You still call grok-4-0709 |
Treat it as a migration issue. Select a current model explicitly and test it; the old slug redirects to Grok 4.3. |
| You need a replacement for that workload | Start by evaluating Grok 4.3, the documented redirect target. Set reasoning effort deliberately rather than relying on the redirect’s low setting. |
| You are starting a new coding or general-work workload | Compare Grok 4.7 with Grok 4.3 using representative tasks, latency, and total cost. |
| You require a specific modality or tool | Verify support in the current documentation for the exact model and API channel before implementation. |
Grok 4.3 capabilities and workload fit
For Grok 4.3, xAI’s model documentation lists text and image input, text output, a 1-million-token context window, function calling, structured outputs, and reasoning settings. Confirm the relevant settings and constraints for your exact integration; listed capabilities alone do not guarantee task-level performance. (docs.x.ai)
A large context window may help with document-heavy work, but it is not evidence that the model will reliably analyze every long input. Test it with your actual documents, prompts, and expected outputs. If you need server-side tools, verify their availability and any applicable charges separately; the model’s function-calling support does not mean every tool is automatically enabled.
Direct API setup: what to confirm
The xAI API overview demonstrates direct API requests using an API key, the https://api.x.ai/v1 base URL, and the Responses API. Its example uses Grok 4.7. Use the current model identifier and endpoint documented for the model you select rather than copying an old Grok 4 identifier from a third-party article. (docs.x.ai)
Before deployment, confirm the selected model’s identifier, supported inputs, reasoning settings, rate limits, and tool requirements. Keep direct xAI API access separate from third-party platform access; one platform’s key-management arrangements do not establish another’s.
API pricing: Grok 4.3 versus Grok 4.7
The migration notice lists Grok 4.3 at $1.25 per million input tokens and $2.50 per million output tokens after the retired slug redirects. (docs.x.ai) The Grok 4.7 model page lists $2.00 per million input tokens and $6.00 per million output tokens. These are API token rates, not consumer subscription prices. (docs.x.ai)
Worked estimate: assuming standard global rates, no cached-token discount, no server-side tool calls, and that each request qualifies for standard rather than long-context pricing, 100 requests of 10,000 input tokens and 2,000 output tokens each use 1 million input tokens and 200,000 output tokens in aggregate. At Grok 4.3’s listed standard rates, that totals about $1.75: $1.25 + (0.2 × $2.50). This is an aggregate across 100 individually short requests—not one 1-million-token request.
Long-context pricing can differ, but the source details available here do not establish Grok 4.3’s exact long-context threshold or rates. Likewise, the pricing details reviewed here do not establish whether server-side tool calls incur separate charges. Verify both on the current pricing documentation for your selected model and tools before estimating production costs. Do not assume the standard-rate example applies to long-context requests.
How to make the production choice
Run a workload-specific evaluation before selecting a model:
- Use representative tasks. Include normal cases, edge cases, and the longest inputs you expect in production.
- Check output quality against your requirements. For structured responses or tool use, validate format compliance and the complete workflow—not just a fluent answer.
- Measure operational behavior. Record latency, errors, and how often outputs need retries or human correction.
- Estimate total cost. Count input and output tokens, and verify long-context pricing, cached-token treatment, regional rates, and tool charges where applicable.
- Pin an explicit current model. Avoid relying on a retired slug’s redirect when you need predictable behavior and billing.
Bottom line: The original Grok 4 API slug grok-4-0709 was retired on May 15, 2026, and now redirects to Grok 4.3 with low reasoning effort. For production, choose and test a current model explicitly. Grok 4.3 is the documented migration target; compare it with Grok 4.7 or another current option when workload quality, cost, or capabilities could change your decision. Which model is best for your application still depends on your own evaluation. (docs.x.ai)