Verdict
Grok Code Fast 1 has positive but anecdotal coding feedback, not evidence that it beats competing models. It was retired from xAI’s API on May 15, 2026. For a new xAI API pilot, use the explicit grok-build-0.1 model ID rather than relying on the legacy name.
xAI’s migration guide says requests to retired model slugs redirect to grok-4.3. In a separate section labeled “Recommended Replacements,” its table recommends grok-build-0.1 for grok-code-fast-1. The current Grok Build 0.1 model page also lists grok-code-fast-1 as an alias. These references do not establish that the legacy slug now routes to Grok Build 0.1; use the explicit ID to select the documented model. Nor do they establish that Grok Build 0.1 preserves the retired model’s performance. xAI migration guide · Grok Build 0.1 model page
For coding quality, a Cursor forum thread includes favorable impressions of “Grok Code,” but does not confirm the exact model version or provide controlled results. xAI’s launch description states intended coding uses, not independent proof of quality. Treat this as a reason to run your own task-based pilot—not a universal model recommendation.
This is a documentation-and-evidence review, not a hands-on model test.
What the evidence says about coding
When xAI introduced Grok Code Fast 1 on August 28, 2025, it described the model as intended for agentic coding and named tools such as grep, a terminal, and file editing. The announcement also highlighted languages including TypeScript, Python, Java, Rust, C++, and Go, and mentioned codebase questions and bug fixes. These are provider statements about intended use, not independent evaluations of accuracy on those tasks. xAI launch announcement
The available user feedback is positive but limited. In a Cursor forum thread dated August 30, 2025, one poster said they preferred “Grok Code” to GPT-5 and Claude Sonnet in their own coding work. Another participant said Grok Code was faster and better than GPT-5 on some tasks, including complex ones. The thread does not identify the exact model version, describe a repeatable task set, or report measured quality scores. Treat it as anecdotal feedback—not a Grok Code Fast 1 benchmark. Cursor forum discussion
A third-party CometAPI article also characterizes Grok Code Fast 1 as coding-focused and suited to agentic workflows. The captured article provides no test results or API specifications, so it is product positioning rather than independent verification. CometAPI article
Practical takeaway: use tasks from your own codebase to test whether the current model works for your team. The evidence here does not establish reliable performance for a particular language, repository, or level of complexity.
Speed, response style, and cost
A Cursor forum poster described “Grok Code” as quick and less conversational between coding steps, and claimed an output rate of “2M tpm.” The thread does not explain how that rate was measured or provide a controlled comparison of completed-task time. It is a user report, not a verified throughput figure.
The same poster said they planned to compare costs later. They described Grok Code as not cheap on a raw per-million-token basis, while speculating that concise output might make it less expensive per task. The captured thread contains no completed comparison, exact prices, or task-normalized cost results. It does not prove a cost advantage. Token speed, raw token price, and cost per correctly completed task are different measures.
Current documented API pricing
xAI’s current pricing documentation lists the direct API rates for grok-build-0.1 below. Prices are in U.S. dollars per million tokens. xAI API pricing · Grok Build 0.1 model page
| Token category | Short context | Long context (prompt reaches 200,000 tokens) |
|---|---|---|
| Input, per 1 million tokens | $1.00 | $2.00 |
| Cached input, per 1 million tokens | $0.20 | $0.40 |
| Output, per 1 million tokens | $2.00 | $4.00 |
When a request’s prompt reaches the model’s 200,000-token long-context threshold, xAI says the long-context rates apply to all tokens in that request. Requests sent to the U.S. regional endpoint are billed at 1.1 times the global token rates, a 10% premium.
For example, across multiple requests whose individual prompts each remain below 200,000 tokens, a combined total of 1 million uncached input tokens and 1 million output tokens would cost $3.00 at global short-context rates. That illustration excludes other applicable charges, such as server-side tool invocations. API charges are separate from editor subscriptions or other platform fees.
These are the documented rates for grok-build-0.1, not evidence of the retired model’s historical price. Because xAI’s documentation discusses both redirects and a recommended replacement for the legacy slug, don’t assume that calling grok-code-fast-1 selects the model or price you intend. Choose an explicit model ID and check actual usage and billing.
Production API fit: resolve the model ID first
The current xAI model page describes grok-build-0.1 as a coding model and lists a 256,000-token context window, function calling, structured outputs, reasoning, and tool use. It also lists grok-code-fast-1 among the model’s aliases. These are documented Grok Build 0.1 details; they should not be read as specifications for the original Grok Code Fast 1 model. Grok Build 0.1 model page
The migration guide says retired model slugs redirect to grok-4.3. Its “Recommended Replacements” table separately assigns grok-build-0.1 to grok-code-fast-1. The model page’s alias list adds another reference to the old name. Taken together, the documents do not make the legacy slug a clear way to select Grok Build 0.1. For a new API pilot, specify grok-build-0.1 directly.
The model page lists request and token rate limits, but the retrieved model details do not establish account-specific limits or how those values apply to a particular deployment. Check the limits shown for your API account. Before production rollout, test the request and tool-calling behavior your application needs, confirm which model responds, and review applicable billing and data-handling terms for your deployment channel. The cited documentation does not settle every operational or data-policy question for a particular application.
Who should pilot it—and how
A short pilot makes sense if your team wants to evaluate a coding-focused model on real work. Don’t adopt it based only on forum impressions, provider descriptions, or a token-price comparison.
Try a small, representative set of tasks, such as:
- A routine bug fix with a clear expected result.
- A codebase question that requires locating and explaining relevant files.
- A bounded feature change that requires editing and running tests.
- A harder task where correctness and review effort matter more than response speed.
Run the same tasks with the alternative model you already use. Record whether results pass tests and review, how much human rework they need, elapsed task time, and actual API cost. Keep prompts, tool access, and acceptance criteria consistent. These are suggested evaluation steps, not results from a test conducted for this review.
One Cursor forum participant said they still used GPT-5 when a task did not work with Grok Code. That is one user’s reported fallback, not evidence that GPT-5 is the right alternative for every team. Retain whichever alternative performs reliably on your own failure cases.
Decision: pilot grok-build-0.1 if it fits your evaluation needs; adopt only if it meets your quality, rework, latency, and cost thresholds. If your integration depends on the exact behavior of the legacy grok-code-fast-1 ID, use the explicit model ID and confirm routing with xAI rather than relying on the legacy alias.
Sources and evidence notes
- xAI launch announcement: intended workflows and language focus for Grok Code Fast 1.
- xAI Grok Build 0.1 model page: current model details, alias, capabilities, context, and listed rate limits.
- xAI migration guide: retirement date, redirect statement, and recommended replacement table.
- xAI API pricing: token rates, long-context billing rule, and U.S. regional endpoint premium.
- Cursor community forum: anecdotal impressions; no exact model-version confirmation or controlled comparison.
- CometAPI article: third-party product positioning, not benchmark or API verification.
Sources
Grok Build 0.1 — xAI API documentation
May 15, 2026 Model Retirement — xAI documentation
API Pricing — xAI documentation