X

xAI Grok 3 Beta

xAI
Vision
131K
Context
$3.00
Input /1M
$15.00
Output /1M
16K
Max Output
Run API Call Open in Chat

xAI Grok 3 Beta

Model Overview

Property Value
Model ID x-ai/grok-3-beta
Provider xAI
Model Type Large Language Model (LLM)
Release Date February 2025 (Beta)
General Availability June 2025
Knowledge Cutoff November 2024

Description

Grok 3 Beta is xAI's flagship reasoning model, described as their "most advanced model" showcasing superior reasoning capabilities and extensive pretraining knowledge. It excels at enterprise use cases including:

  • Data Extraction - Parsing and extracting structured information from documents
  • Coding - Code generation, debugging, and optimization
  • Text Summarization - Condensing large documents while preserving key information

The model demonstrates deep domain knowledge across:

  • Finance
  • Healthcare
  • Law
  • Science

Technical Specifications

Specification Value
Context Length 131,072 tokens
Max Output (Playground) 16,000 tokens per run
Max Output (API) Up to 131,072 tokens (prompt + response combined)
Input Modalities Text
Output Modalities Text
Reasoning Mode No (standard mode)
Function Calling Yes
Structured Outputs Yes
Cached Input Tokens Yes

Training Infrastructure

  • Compute: 10-15x increase over Grok 2
  • Training Cluster: xAI's Colossus supercluster with 100K+ NVIDIA H100 GPUs
  • Memphis Supercomputer: Built in 122 days

Pricing

LangMart Pricing

Type Cost per Million
Input $3.00
Output $15.00
Input Cache Read $0.75

AI/ML API Pricing

Type Cost per Million
Input $3.15 - $4.00
Output $15.75 - $20.00
Cached $0.75 - $1.00

Model Variants Comparison

Model Input (per 1M) Output (per 1M) Notes
Grok 3 $3.00 $15.00 Standard model
Grok 3 Fast $5.00 $25.00 Lower latency
Grok 3 Mini $0.30 $0.50 Lightweight variant

Capabilities

Core Capabilities

  • Advanced multistep reasoning with self-correcting error detection
  • Mathematical problem-solving
  • Code generation and debugging
  • Knowledge retrieval with internet information integration
  • Strong instruction-following ability

Tool Support

  • Function calling enabled
  • Structured outputs supported
  • Tool choice options: none, auto, required, function
  • Maximum 128 functions per request

Special Features

  • Deep Search: Information retrieval capability
  • Big Brain Mode: Enhanced processing for complex problems
  • Think Mode: Extended reasoning for difficult tasks

Supported Parameters

Required Parameters

Parameter Type Description
model string Must be x-ai/grok-3-beta
messages array Array of message objects for the conversation

Optional Parameters

Parameter Type Range Description
max_tokens number min: 1 Maximum tokens for chat completion
max_completion_tokens integer min: 1 Upper bound for generated tokens
temperature number 0-2 Sampling temperature for output randomness
top_p number 0.01-1 Nucleus sampling parameter
top_k number - Sample from top K options only
seed integer min: 1 For deterministic sampling (Beta)
frequency_penalty number -2 to 2 Reduce likelihood of repeated sequences
presence_penalty number -2 to 2 Penalize tokens based on prior appearance
stop array max: 4 Sequences where generation stops
logprobs boolean - Return output token log probabilities
top_logprobs integer - Number of top logprobs to return
stream boolean - Enable server-sent event streaming
response_format object - Specify JSON or structured output formats
tools array max: 128 Available functions for model calls
tool_choice string/object - Controls tool invocation behavior

Response Format Options

  • Text: Default plain text response
  • JSON Object: Legacy JSON mode
  • JSON Schema: Structured JSON responses with schema validation (recommended)

Limitations

Known Weaknesses

  • Complex coding tasks may still have gaps
  • Symbolic logic limitations
  • Limited humor and creativity compared to some competitors
  • Citation hallucinations reported
  • No image/multimodal input support (text-only)

Rate Limits

  • 100,000 tokens per minute (TPM) per customer/tenancy
  • Request limit increases available through provider support

xAI Model Family

Model Context Input Cost Output Cost Use Case
x-ai/grok-3 131K $3/M $15/M Standard flagship
x-ai/grok-3-fast 131K $5/M $25/M Low latency
x-ai/grok-3-mini 131K $0.30/M $0.50/M Cost-effective
x-ai/grok-3-mini-beta 131K - - Mini beta variant
x-ai/grok-4 2M - - Next generation
x-ai/grok-4-fast 2M $0.20/M $0.50/M Fast multimodal
x-ai/grok-4.1-fast 2M $0.20/M $0.50/M Best agentic model

Thinking/Reasoning Variants

Model Description
Grok 3 Beta (Think) Extended reasoning mode
Grok 3 Reasoning Beta Dedicated reasoning variant
Grok 3 Mini Reasoning Lightweight reasoning

Benchmark Performance

Mathematics (AIME)

Benchmark Score Notes
AIME 2025 93.3% Solved fresh, unseen problems
AIME 2025 (single attempt) 86.7% Without consensus sampling
AIME 2024 (single attempt) 92.0% Single attempt score
AIME 2024 (Think Mode) 95.8% With extended reasoning

Coding

Benchmark Score
LiveCodeBench 79.4%
LiveCodeBench (Think Mode) 80.4%
SWE-Bench Verified 63.8%

Science & Knowledge

Benchmark Score
GPQA (Graduate-Level Q&A) 84.6%
MMLU-Pro 79.9%

LMArena Rankings

  • First model to break the 1400 ELO score barrier
  • Ranks #1 across all categories:
    • Overall performance
    • Hard Prompts
    • Coding
    • Math
    • Creative Writing
    • Instruction Following
    • Longer Query Handling
    • Multi-Turn Conversations

Comparative Performance

Grok 3 Beta outperforms:

  • GPT-4o
  • Claude 3.5 Sonnet
  • Gemini 2.0 Pro
  • DeepSeek V3

Note: Still trails behind OpenAI's o3 model in some benchmarks.

Provider Information

OpenRouter

Property Value
Base URL https://api.langmart.ai/v1
Data Retention 30 days
Training on User Data No
Publishing Rights Not permitted
Requires User IDs Yes

xAI Direct API

Property Value
Base URL https://api.x.ai/v1
API Compatibility OpenAI SDK compatible
Rate Limit 100,000 TPM per customer

API Usage Examples

Basic Chat Completion

curl -X POST https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "x-ai/grok-3-beta",
    "messages": [
      {"role": "user", "content": "Explain quantum computing in simple terms"}
    ],
    "temperature": 0.7,
    "max_tokens": 1000
  }'

With Function Calling

curl -X POST https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "x-ai/grok-3-beta",
    "messages": [
      {"role": "user", "content": "What is the weather in San Francisco?"}
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_weather",
          "description": "Get weather for a location",
          "parameters": {
            "type": "object",
            "properties": {
              "location": {"type": "string"}
            },
            "required": ["location"]
          }
        }
      }
    ],
    "tool_choice": "auto"
  }'

Structured Output (JSON Schema)

curl -X POST https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "x-ai/grok-3-beta",
    "messages": [
      {"role": "user", "content": "Extract person info: John Doe, age 30, engineer"}
    ],
    "response_format": {
      "type": "json_schema",
      "json_schema": {
        "name": "person",
        "schema": {
          "type": "object",
          "properties": {
            "name": {"type": "string"},
            "age": {"type": "integer"},
            "occupation": {"type": "string"}
          },
          "required": ["name", "age", "occupation"]
        }
      }
    }
  }'

Data Policy

Policy Value
Training on Prompts No
Prompt Retention 30 days
Output Publishing Not permitted
User ID Required Yes
Terms of Service https://x.ai/legal/terms-of-service-enterprise

Changelog

Date Event
February 2025 Grok 3 Beta announced
May 22, 2025 Oracle Cloud beta release
June 10, 2025 OpenRouter availability
June 24, 2025 General availability

Sources