Short answer
As of November 9, 2026, don’t assume grok-code-fast-1 is a new production API option without resolving xAI’s conflicting migration documentation. The retirement notice says requests to the retired model route to grok-4.3 after May 15, 2026, but its replacement table says grok-build-0.1. The model page identifies Grok Build 0.1 and lists grok-code-fast-1 as an alias. The available excerpts do not resolve what the old name serves today. Confirm the explicit model ID in your xAI account before deployment; don’t assume the legacy name preserves the original model’s behavior or price. (xAI migration notice)
The available model-page excerpt establishes the name grok-build-0.1 and the legacy aliases, but does not establish the model’s current coding capabilities or practical limits. Older descriptions of Grok Code Fast 1 as a model for agentic coding are not enough to confirm the successor’s features or performance on your workload.
API access: distinguish the documented pattern from a model-specific call
The captured xAI quickstart documents a Responses API request pattern using bearer authentication, a model field, and an input field—but its example uses grok-4.7. That example does not confirm that grok-build-0.1 accepts the same request. The available documentation excerpts also do not establish how to extract generated text and token usage from a response for this model.
For that reason, this guide does not present a model-specific production request as verified. Before integrating, confirm in current xAI documentation or your account that the exact model ID supports the intended endpoint and request fields, and check the documented response structure for generated text and usage. The available excerpts do not establish those details for grok-build-0.1. They also do not establish its current rate limits; check the limits for your account and selected model rather than relying on an assumed team-tier rule. (xAI API quickstart · xAI rate limits)
Current listed API pricing
The xAI pricing page lists USD per 1 million tokens for grok-build-0.1:
| Context length | Input | Cached input | Output |
|---|---|---|---|
| Short context | $1.00 | $0.20 | $2.00 |
| Long context (at least 200,000 tokens) | $2.00 | $0.40 | $4.00 |
The pricing documentation says that once a request reaches the long-context threshold, all tokens in that request are billed at the long-context rates. It also says requests to the US regional endpoint cost 1.1 times the global token rates. These are the listed rates for grok-build-0.1, not a guarantee of the price or behavior you will receive by sending a retired alias. Confirm the selected model and rates in your account before forecasting production spend. (xAI API pricing)
For illustration, a short-context request using 10,000 input tokens and 2,000 output tokens would cost $0.014 at those listed standard rates: 10,000 × $1.00 / 1,000,000, plus 2,000 × $2.00 / 1,000,000. This assumes short-context rates and no cached-input discount. (xAI API pricing)
Older third-party articles reported $0.20 per million input tokens and $1.50 per million output tokens. Those figures do not match xAI’s listed grok-build-0.1 rates, so don’t use them for a current API estimate. IDE offers and third-party gateway billing are separate from direct xAI API pricing.
What older coding benchmarks can—and can’t—tell you
Vals AI’s August 27, 2025 evaluation reported Grok Code Fast at 62% on LCB, 4.3% on IOI v1 (8th of 12), and 57.6% on SWE-bench Verified. In its SWE-bench comparison, Vals AI reported Grok 4 at 58.6%; it gave latency figures of 264.68 seconds for Grok Code Fast and 704.8 seconds for Grok 4. The captured evaluation does not specify its test configuration or how it measured latency. Treat these as attributed historical results, not production guarantees or a current comparison of the successor model. (Vals AI evaluation)
Another third-party article reports approximately 70.8% on “SWE-Bench,” but the captured text does not identify enough benchmark details to reconcile that figure with Vals AI’s SWE-bench Verified result. Don’t combine the scores or treat either as a like-for-like current ranking. (CodeGPT article)
Test the model ID confirmed for your account on representative tasks from your own work: code edits with tests, multi-step tool use, bug fixes with edge cases, and repository context. Compare correctness and test pass rate alongside latency, token use, and cost. Include failure cases and human review for consequential changes.
Bottom line
Grok Code Fast 1 was positioned for agentic coding, but the API model was retired on May 15, 2026. The available xAI excerpts identify Grok Build 0.1 and list its rates, but they do not resolve the old slug’s routing conflict or establish a model-specific API call pattern and coding feature set. For a new integration, verify the explicit model ID, endpoint compatibility, response fields, limits, and current price in your xAI account and documentation, then evaluate that model on your coding workload before adoption.
Sources
Grok Code Fast 1 / Grok Build 0.1 model details
Grok Code Fast Benchmarks, Cost and Capabilities
The Rise of Grok Code Fast 1: An Analysis of Market Dominance in AI Coding