Skip to main content

​
Available Models

/grok-x supports three Grok-4 models. Choose based on latency, depth, and cost requirements:
FlagModelUse when
(none)grok-4-1-fast-reasoningDefault — complex analysis
--fastgrok-4-1-fast-non-reasoningQuick lookups, simple searches
--deepgrok-4-0709Maximum depth, higher cost
--extractgrok-4-1-fast-non-reasoningAuto-selected — required for structured output with tools
You can set a custom default model using the GROK_X_DEFAULT_MODEL environment variable in ~/.claude/settings.json. If omitted, the skill defaults to grok-4-1-fast-reasoning.

​
Model Comparison

Default

​
grok-4-1-fast-reasoning

The balanced choice. Uses extended reasoning to produce thorough, well-organized analysis.Best for:
  • Open-ended research
  • Complex questions
  • Multi-step analysis
  • --web and --analyze modes
Characteristics:
  • Moderate latency
  • High quality
  • Good cost/performance ratio
Example:
/grok-x "What is happening with AI regulation on X?"

Fast

​
grok-4-1-fast-non-reasoning

The speed choice. Non-reasoning variant with significantly lower latency and cost.Best for:
  • Simple lookups
  • Quick quote pulls
  • Time-sensitive queries
  • --extract mode (auto-selected)
Characteristics:
  • Lowest latency
  • Lower cost
  • No extended reasoning chain
Example:
/grok-x "What did @naval post today?" --fast

Deep

​
grok-4-0709

The flagship choice. Maximum analytical depth for complex topics.Best for:
  • Geopolitical analysis
  • Dense technical topics
  • When reasoning chain matters
  • High-stakes research
Characteristics:
  • Highest latency
  • Highest cost
  • Maximum depth
Example:
/grok-x "Comprehensive analysis of the tariff debate on X" --deep --web --images --citations

​
When to Use Each Model

​
Default: grok-4-1-fast-reasoning

Use this when:
  • Quality matters more than speed
  • The query requires multi-step reasoning
  • You’re combining tools (--web, --analyze)
  • You need well-structured, thorough analysis
Examples: 
# Cross-platform research
/grok-x "Are there signs of a chip export control reversal?" --web --citations

# Data analysis
/grok-x "Tech layoff announcements this month" --analyze --from 2026-02-01

# Handle monitoring with context
/grok-x "What has @atrupar posted about this week?" --handles atrupar --from 2026-02-25
Output quality: High. The model uses extended reasoning to organize findings, identify patterns, and structure analysis.

​
Fast: grok-4-1-fast-non-reasoning (with --fast)

Use this when:
  • You need results in seconds, not tens of seconds
  • The query is straightforward
  • You’re doing quick lookups or quote pulls
  • Cost optimization is a priority
Examples: 
# Quick lookup
/grok-x "What did @naval post today?" --fast

# Simple quote pull
/grok-x "Latest @elonmusk tweet about SpaceX" --fast

# Fast handle monitoring
/grok-x "@pmarca recent posts" --handles pmarca --fast
Output quality: Good for simple queries. Less structured than default, but significantly faster.
The --fast flag is also auto-selected when using --extract. Structured output with tools requires a non-reasoning model.

​
Deep: grok-4-0709 (with --deep)

Use this when:
  • Maximum analytical depth is required
  • The topic is complex (geopolitics, dense technical subjects)
  • The reasoning chain itself matters
  • You’re willing to pay higher cost and wait longer
Examples: 
# Complex geopolitical analysis
/grok-x "Comprehensive analysis of the tariff debate on X" --deep --web --images --citations

# Dense technical topic
/grok-x "xAI Responses API developer feedback" --deep --web

# High-stakes research
/grok-x "AI safety discourse" --deep --handles karpathy,ylecun,sama
Output quality: Maximum. The flagship model delivers the most thorough reasoning chains and deepest analysis.
Higher cost and latency. Only use --deep when the default model isn’t sufficient. For routine searches, stick with the default.

​
Automatic Model Selection: --extract

The --extract flag automatically selects grok-4-1-fast-non-reasoning, regardless of --fast or --deep. Why? Structured output with tools requires a non-reasoning Grok-4 variant. This is an xAI API constraint. Examples: 
# Sentiment extraction (auto-selects fast non-reasoning)
/grok-x "Bitcoin sentiment on X right now" --extract sentiment

# Timeline extraction (auto-selects fast non-reasoning)
/grok-x "xAI product announcements in the last 90 days" --extract timeline --from 2025-12-01

# Claims extraction (auto-selects fast non-reasoning)
/grok-x "Claims being made about AGI on X" --extract claims --from 2025-01-01 --web
You cannot override this behavior. --extract always uses grok-4-1-fast-non-reasoning for schema compliance.

​
Setting a Custom Default

You can override the default model (grok-4-1-fast-reasoning) by setting GROK_X_DEFAULT_MODEL in ~/.claude/settings.json: 
{
  "env": {
    "XAI_API_KEY": "xai-your-key-here",
    "GROK_X_DEFAULT_MODEL": "grok-4-1-fast-non-reasoning"
  }
}
Valid values:
  • grok-4-1-fast-reasoning (default)
  • grok-4-1-fast-non-reasoning (fast)
  • grok-4-0709 (deep)
The --fast and --deep flags always override the default, regardless of GROK_X_DEFAULT_MODEL.

​
Decision Guide

Quick reference:
  • Simple lookup? → --fast
  • Complex analysis? → default (no flag)
  • Maximum depth? → --deep
  • Structured output? → --extract (auto-selects fast)

​
Cost Considerations

Model costs are determined by xAI’s pricing. As of March 2026:
  • grok-4-1-fast-non-reasoning is the most cost-effective
  • grok-4-1-fast-reasoning is moderately priced
  • grok-4-0709 is the most expensive
Check console.x.ai/pricing for current rates.
Optimization tips:
  1. Use --fast for routine queries to reduce costs
  2. Reserve --deep for high-value analysis
  3. Combine --fast with date filters (--from, --to) to limit search scope
  4. Use --extract when you need structured output — it’s faster and cheaper than --analyze

​
Next Steps

Tool Combinations

Learn how to combine models with different tool modes for optimal results

Build docs developers (and LLMs) love

Get started for freeTalk to us