Back to models
Chat Models

xAI

Grok 4.20 Reasoning

Best for Research

Reasoning-enabled Grok 4.20 with 1M context, image input, tool calling, and structured outputs.

Reasoning1M context
Open AppBrowse All Models

At a glance

Know the model before you prompt.

xAI API specifications
Context window
1,000,000 tokens
Maximum output
Not verified
Inputs → output
Text + Images → Text
Knowledge cutoff
Not verified

API model ID: grok-4.20-0309-reasoning

Capabilities & boundaries

What it supports. Where the limits are.

Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.

Supported API features and tools

  • Reasoning-enabled responses for multi-step tasks
  • A 1,000,000-token context with text and image input
  • Function calling for connected tools
  • Structured outputs for organized responses
  • Batch API support with a published 20% token discount

Before you choose

  • The card confirms reasoning but does not list configurable effort levels or a default. Do not copy Multi-Agent's agent-count settings.
  • A separate maximum output limit and training cutoff are not verified in the reviewed model documentation.
  • Large context and provider claims about reliability do not remove the need to check conclusions, citations and tool results.
  • This is a text-output model. Search and other actions require enabled tools; API capabilities are not proof of app integration.

Reasoning behavior

This is the reasoning-enabled 0309 variant. The reviewed card does not establish a configurable effort list or default. Use the Non-Reasoning variant when evaluating a no-reasoning alternative rather than assuming an undocumented none setting switches this model's behavior.

  • Use grok-4.20-0309-reasoning to identify the documented variant. The card also lists grok-4.20 and grok-4.20-reasoning aliases.
  • Do not confuse the website route grok-4-20 with an API model ID. The route is preserved for existing links; the API identifier is shown in Specifications.
  • Batch support and tool support are separate capabilities. Verify the exact request and endpoint rather than assuming every tool is available in every processing mode.

Put it to work

Start with a more useful prompt.

Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.

Workflow 01

Reconcile competing constraints

Turn a complex planning problem into an explicit trade-off decision.

Evaluate this project plan against the budget, deadline and staffing constraints below. Identify which combinations are feasible, show the arithmetic behind each option, and call out assumptions. Recommend the smallest scope adjustment needed for a feasible plan. Separate hard constraints from preferences and list the evidence that would change your recommendation.

Workflow 02

Compare incident hypotheses

Distinguish causal evidence from symptoms in a technical incident.

Using the incident timeline, logs and recent changes below, compare the proposed root-cause hypotheses. For each, list supporting evidence, contradictions and one low-risk diagnostic check. Rank them by evidential support rather than narrative plausibility. Do not claim causation or recommend destructive action without sufficient evidence.

Workflow 03

Stress-test a decision

Make an existing recommendation easier to challenge and improve.

Review this decision memo and its supporting evidence. Reconstruct the strongest case for the recommendation, then identify assumptions whose failure would reverse it. Compare two alternatives using the same criteria. Finish with a risk register and concrete validation steps, citing the supplied sections rather than inventing external facts.

Developer reference

xAI API pricing

These are global xAI API reference prices, not EZ Ai Assist subscription prices.

View EZ Ai Assist plans
Standard processing · USD per 1,000,000 tokens
Token type< 200,000prompt tokens≥ 200,000prompt tokens
Input$1.25$2.50
Cached input$0.20$0.40
Output$2.50$5.00
Additional API fees · USD
UsageRate and unit
Web Search$5.00per 1,000 calls
X Search: posts$5.00per 1,000 posts fetched
X Search: profiles$10.00per 1,000 profiles fetched
Code execution$5.00per 1,000 calls
  • At 200,000 prompt tokens or more, the higher token rates apply to the entire request, not just excess input.
  • Reasoning tokens contribute to billed usage. Equal token rates across the 4.20 variants do not imply equal total cost per answer.
  • The model supports Batch API with a 20% token discount. Priority processing uses 2× rates when the response confirms priority; the table shows standard global rates.
  • Enabled tools add charges. Web Search and code execution are call-based; X Search is billed per fetched post or profile. Check actual usage and integration support.

Common questions

A few things worth knowing.

Which Grok 4.20 variant is this?

This guide covers grok-4.20-0309-reasoning. It retains the existing grok-4-20 website URL. Non-Reasoning and Multi-Agent are separate catalog pages and API variants, not interchangeable names for this guide.

What is its context window?

The model card lists 1,000,000 tokens of context. It does not establish a separate output ceiling or training cutoff in the reviewed material. Context capacity should not be confused with allowed output length or app limits.

Which reasoning levels should I send?

The reviewed card confirms reasoning but does not establish a configurable effort list or default. Do not copy Grok 4.7's levels or Multi-Agent's agent-count mapping. Confirm current API behavior before sending a setting.

Can it inspect images and call tools?

Yes, the card lists text/image input, text output, function calling and structured outputs. Tool use needs an enabled integration and appropriate permissions. Inspect proposed actions and validate returned data before relying on them.

How is it different from Non-Reasoning?

This variant reasons before producing its answer; Non-Reasoning is documented without that capability. Use a representative task set to compare accuracy, latency and total usage. Identical published token rates do not mean identical request costs.

Does it support Batch and long-context pricing?

Yes. The card lists a 20% Batch token discount and higher standard rates starting at 200,000 prompt tokens. The higher rates apply to the whole request, including output. Enabled tools can incur separate charges.

Are these xAI rates the EZ Ai Assist subscription price?

No. They are developer reference rates. Check EZ Ai Assist for plan prices, model access, tool availability and limits; this guide does not establish that every provider setting is exposed in the app.

Check the source

Official documentation

Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.

Same provider

More from xAI

xAI

Grok 4.7

Coding and knowledge-work model with 500K context, image input, and configurable reasoning effort.

xAI

Grok 4.6

Coding and agentic model with 500K context, structured outputs, and four reasoning effort levels.

xAI

Grok 4.5

Coding and engineering model with 500K context, image input, tool calling, and configurable reasoning.

xAI

Grok 4.3

1M-context model for instruction-driven work and tool calling, with configurable reasoning that can be disabled.