Choose a Grok identifier by the task, supported inputs and operating conditions. This catalog keeps live model specifications close to the lifecycle and evaluation questions needed for a controlled application change.

Model families at xAI

The model catalog spans Grok language models and Imagine media generation, with Voice documented through its own service surface. Match the requested output to the correct endpoint before comparing models. Official documentation.

Image understanding and image generation are different capabilities. A model that can inspect a supplied image is not automatically the model that produces a new image. Check accepted inputs, response types and billing units for the exact operation you plan to implement.

Server-side tools can add web or X retrieval to supported workflows. Fresh retrieval should be an explicit integration choice with inspected citations, rather than a property inferred from a model’s name. Official documentation.

For a candidate shortlist, start with hard requirements: input modality, output structure, tool interface and context needs. Then evaluate task quality on examples that expose likely failures. Avoid calling the largest catalog entry the strongest without comparable evidence.

All models

Model catalog
ModelOfficial identifierModalitiesContext tokensInput USD / 1MOutput USD / 1MStatus
grok-4.20-0309-non-reasoninggrok-4.20-0309-non-reasoningtext, image1,000,000$1.25$2.5Active
grok-4.20-0309-reasoninggrok-4.20-0309-reasoningtext, image1,000,000$1.25$2.5Active
grok-4.20-multi-agent-0309grok-4.20-multi-agent-0309text, image1,000,000$1.25$2.5Active
grok-4.3grok-4.3text, image1,000,000$1.25$2.5Active
grok-4.5grok-4.5text, image500,000$2$6Active
grok-4.6grok-4.6text, image500,000$2$6Active
grok-build-0.1grok-build-0.1text, image256,000$1$2Active

Last verified · Source ↗

Use the catalog’s exact identifiers and capability notes. The models reference also explains aliases, while dated identifiers can provide a more specific target for reproducible deployment. Official documentation.

Keep the chosen identifier in deployment configuration. Store the evaluation results alongside it so a future operator can see why it was selected. If a field is unverified, check its official source before making it an application requirement.

Cheapest / strongest / longest-context

Only active candidates with documented values for this ranking are included. Prices retain the tier and deployment condition shown below.

Models ranked by input price
ModelProviderInput USD / 1MOutput USD / 1MContext tokensPrice conditionCost
grok-build-0.1xAI$1$2256,000short contextEstimate cost
grok-4.20-0309-non-reasoningxAI$1.25$2.51,000,000short contextEstimate cost
grok-4.20-0309-reasoningxAI$1.25$2.51,000,000short contextEstimate cost
grok-4.20-multi-agent-0309xAI$1.25$2.51,000,000short contextEstimate cost
grok-4.3xAI$1.25$2.51,000,000short contextEstimate cost
grok-4.5xAI$2$6500,000short contextEstimate cost
grok-4.6xAI$2$6500,000short contextEstimate cost

Last verified · Source ↗

Only active candidates with documented values for this ranking are included. Prices retain the tier and deployment condition shown below.

Models ranked by context
ModelProviderInput USD / 1MOutput USD / 1MCached input USD / 1MContext tokensPrice conditionCost
grok-4.20-0309-non-reasoningxAI$1.25$2.5$0.21,000,000short contextEstimate cost
grok-4.20-0309-reasoningxAI$1.25$2.5$0.21,000,000short contextEstimate cost
grok-4.20-multi-agent-0309xAI$1.25$2.5$0.21,000,000short contextEstimate cost
grok-4.3xAI$1.25$2.5$0.21,000,000short contextEstimate cost
grok-4.5xAI$2$6$0.3500,000short contextEstimate cost
grok-4.6xAI$2$6$0.5500,000short contextEstimate cost
grok-build-0.1xAI$1$2$0.2256,000short contextEstimate cost

Last verified · Source ↗

These rankings describe numerical properties, not universal quality. Compare a candidate on the complete task: whether it uses evidence correctly, follows required structure and produces an output your application can validate.

For long inputs, test retrieval and prompt assembly separately. A large supported window does not tell you whether the relevant detail will be used reliably in every task. Measure whether a focused source selection improves the answer before making larger prompts the default.

The applicable pricing condition can change with prompt length. Compare the selected candidates under the tier that matches the actual assembled request. Official documentation.

Deprecations and replacements

Review release notes and migration guidance before replacing a deployed model or endpoint. A retirement notice has a timing and scope; a future announced change should not be treated as already effective. Official documentation.

Keep a migration record containing the old and new identifiers, request-shape changes and the saved tasks that were reevaluated. If rollback is still possible, preserve that configuration while the new candidate is observed under real workload variation.

A successful response is only the first migration check. Confirm that structured fields, citations, tool arguments and completion handling still satisfy the application. Changes in timing or token usage also need to be reflected in operational budgets.

Naming and versions

xAI distinguishes family aliases, latest aliases and versioned model identifiers. Choose the naming behavior intentionally: an alias can simplify updates, while a specific version makes the deployed target more explicit. Official documentation.

Do not construct a guessed identifier from a marketing name or copy a retired example without checking it. Keep model and endpoint validation separate; a valid model sent to an incorrect route still fails.

Use xAI key permissions if a listed model is inaccessible, error diagnosis for the actual rejection and the xAI hub for the complete operating reference.

Use the AI API cost calculator to turn the model and workload you are considering into an estimate.

Last verified · Source ↗

Frequently asked questions

Does an image-capable text model generate images?
Not necessarily. Image input and image output are distinct capabilities; verify the endpoint and selected model.
Does a model automatically retrieve current information?
Use the documented search tools when retrieval is needed and inspect the returned evidence. Official documentation.
What does the context ranking prove?
It compares documented capacity. It does not establish answer quality or evidence use on a particular application.
Should I use an alias or a dated identifier?
Choose based on the update behavior you need and the provider’s documented naming contract. Keep the decision with your deployment configuration. Official documentation.
What should be checked after a model replacement?
Rerun difficult saved tasks, validate output structure and tool arguments, and compare timing, usage and failure handling.

Sources

Last verified · Source ↗