Unlock exclusive introductory pricing for the newly launched Gemini 3.5 Flash.
X

Grok 4.20

مدخل:$1/M
الإخراج:$2/M
سياق:2,000,000
تاريخ الإصدار:Mar 11, 2026

Grok 4.20 release introduces a multi-agent architecture (multiple specialized agents coordinated in real time), expanded context modes, and focused improvements to instruction-following, hallucination reduction, and structured/tooled outputs.

جديد
الاستخدام التجاري

Playground لـ Grok 4.20

استكشف Playground الخاص بـ Grok 4.20 — بيئة تفاعلية لاختبار النماذج وتشغيل الاستعلامات في الوقت الفعلي. جرب المطالبات، وضبط المعاملات، والتكرار فورًا لتسريع التطوير والتحقق من حالات الاستخدام.

Technical Specifications of Grok-4.20

ItemGrok-4.20 (public specs)
Model familyGrok-4 series
DeveloperxAI
Release statusBeta (first rollout Feb 17, 2026)
Input typesText, Image, Video
Output typesText outputs (structured outputs & function/tool calling supported).
Context windowUp to 2,000,000 tokens
ArchitectureMulti-agent collaborative reasoning
Tool supportFunction calling, structured outputs
ReasoningBuilt-in reasoning capabilities
Training infrastructureColossus supercluster (~200,000 GPUs)
Model variantsgrok-4.20-multi-agent-beta-0309, grok-4.20-beta-0309-reasoning, grok-4.20-beta-0309-non-reasoning.

What is Grok-4.20

Grok-4.20 is the latest experimental release in the Grok-4 family developed by xAI. It focuses on agentic reasoning, extremely long context handling, and high-speed inference, aiming to deliver precise answers with a lower hallucination rate than earlier Grok models.

Unlike earlier Grok models that used single-model inference, Grok-4.20 introduces multi-agent collaboration, where several internal agents analyze a prompt simultaneously and converge on a final answer. This architecture is designed to improve performance on complex reasoning, coding, and research tasks.

Main Features of Grok-4.20

  • Ultra-long context window (2M tokens): Enables processing of entire books, large datasets, or long coding repositories in a single prompt.
  • Multi-agent reasoning architecture: Up to four internal agents can analyze a prompt in parallel and debate solutions before producing a final answer.
  • Agentic tool calling and structured outputs: Supports function calling and structured responses for integration with applications and automated workflows.
  • Multimodal understanding: Accepts text, image, and video inputs within the same model pipeline.
  • Fast inference with low hallucination focus: xAI positions the model as optimized for truthful answers and strong prompt adherence.

Benchmark Performance of Grok-4.20

Public benchmark data is still limited during beta, but early reports indicate:

BenchmarkResult / Status
LMSYS Chatbot ArenaEstimated ELO ~1505–1535
ForecastBenchRanked #2 in early tests
Alpha Arena trading challengeAchieved +34.59% returns

These numbers suggest Grok-4.20 competes with frontier models in real-world reasoning and agent-driven tasks rather than simple benchmark questions.

Grok-4.20 Beta vs Other Frontier Models

ModelDeveloperContext WindowKey Strength
Grok-4.20xAI2M tokensMulti-agent reasoning
GPT-5.2OpenAI~400K tokensAdvanced reasoning + coding
Gemini 3 ProGoogle~1M tokensmultimodal and Google ecosystem
Claude 4 OpusAnthropic~200K+ tokensreliable reasoning

Key differences

  • Grok-4.20 emphasizes multi-agent collaboration for reasoning tasks.
  • It provides one of the largest context windows in production LLMs (2M tokens).
  • Competing models may outperform Grok in certain areas such as structured reasoning or creative writing depending on evaluation tasks.

Representative Use Cases

  1. Long-context research analysis
    Process large documents, legal materials, or academic research.
  2. Agentic automation systems
    Build multi-step workflows where the model plans and executes tasks.
  3. Advanced coding and simulations
    Solve engineering problems or simulate systems with long reasoning chains.
  4. Data analysis and dashboard automation
    Track and analyze multiple streams of data in parallel.
  5. Multimodal knowledge processing
    Interpret images, video frames, and text in a unified reasoning process.

How to access and use Grok 4.2 API

Step 1: Sign Up for API Key

Log in to cometapi.com. If you are not our user yet, please register first. Sign into your CometAPI console. Get the access credential API key of the interface. Click “Add Token” at the API token in the personal center, get the token key: sk-xxxxx and submit.

Step 2: Send Requests to Grok 4.2 API

Select the “grok-4.20-0309-reasoning” endpoint to send the API request and set the request body. The request method and request body are obtained from our website API doc. Our website also provides Apifox test for your convenience. Replace <YOUR_API_KEY> with your actual CometAPI key from your account. Where to call it:   Chat format.

Insert your question or request into the content field—this is what the model will respond to . Process the API response to get the generated answer.

Step 3: Retrieve and Verify Results

Process the API response to get the generated answer. After processing, the API responds with the task status and output data.

الأسئلة الشائعة

التسعير لـ Grok 4.20

استكشف الأسعار التنافسية لـ Grok 4.20، المصمم ليناسب الميزانيات المختلفة واحتياجات الاستخدام المتنوعة. تضمن خططنا المرنة أن تدفع فقط مقابل ما تستخدمه، مما يجعل من السهل التوسع مع نمو متطلباتك. اكتشف كيف يمكن لـ Grok 4.20 تحسين مشاريعك مع الحفاظ على التكاليف قابلة للإدارة.

ModelComet Price (USD / M Tokens)Official Price (USD / M Tokens)Discount
grok-4.20-0309-reasoning
مدخل:$1/M
الإخراج:$2/M
مدخل:$1.25/M
الإخراج:$2.5/M
-20%

نموذج الكود وواجهة برمجة التطبيقات لـ Grok 4.20

احصل على أكواد نماذجية شاملة وموارد API لـ Grok 4.20 لتبسيط عملية التكامل الخاصة بك. توفر وثائقنا التفصيلية إرشادات خطوة بخطوة، مما يساعدك على الاستفادة من الإمكانات الكاملة لـ Grok 4.20 في مشاريعك.

#!/usr/bin/env bash
# Get your CometAPI key from https://api.cometapi.com/console/token
# Export it as: export COMETAPI_KEY="your-key-here"

response=$(curl --silent --location --request POST "https://api.cometapi.com/v1/responses" \
  --header "Authorization: Bearer $COMETAPI_KEY" \
  --header "Content-Type: application/json" \
  --header "Accept: application/json" \
  --data-raw '{
    "model": "grok-4.20-multi-agent-beta-0309",
    "input": [
      {
        "role": "user",
        "content": "Research the latest breakthroughs in quantum computing and summarize the key findings."
      }
    ],
    "tools": [
      {"type": "web_search"},
      {"type": "x_search"}
    ]
  }')

if command -v jq >/dev/null 2>&1; then
  printf '%s
' "$response" | jq -r '(
    [
      .output[]?
      | select(.type == "message")
      | .content[]?
      | select(.type == "output_text")
      | .text
    ][0]
  ) // .output_text // .'
else
  printf '%s
' "$response"
fi

cURL Code Example

#!/usr/bin/env bash
# Get your CometAPI key from https://api.cometapi.com/console/token
# Export it as: export COMETAPI_KEY="your-key-here"

response=$(curl --silent --location --request POST "https://api.cometapi.com/v1/responses" \
  --header "Authorization: Bearer $COMETAPI_KEY" \
  --header "Content-Type: application/json" \
  --header "Accept: application/json" \
  --data-raw '{
    "model": "grok-4.20-multi-agent-beta-0309",
    "input": [
      {
        "role": "user",
        "content": "Research the latest breakthroughs in quantum computing and summarize the key findings."
      }
    ],
    "tools": [
      {"type": "web_search"},
      {"type": "x_search"}
    ]
  }')

if command -v jq >/dev/null 2>&1; then
  printf '%s\n' "$response" | jq -r '(
    [
      .output[]?
      | select(.type == "message")
      | .content[]?
      | select(.type == "output_text")
      | .text
    ][0]
  ) // .output_text // .'
else
  printf '%s\n' "$response"
fi

Python Code Example

import os

from openai import OpenAI

# Get your CometAPI key from https://api.cometapi.com/console/token, and paste it here
COMETAPI_KEY = os.environ.get("COMETAPI_KEY") or "<YOUR_COMETAPI_KEY>"
BASE_URL = "https://api.cometapi.com/v1"

client = OpenAI(base_url=BASE_URL, api_key=COMETAPI_KEY)
response = client.responses.create(
    model="grok-4.20-multi-agent-beta-0309",
    input=[
        {
            "role": "user",
            "content": "Research the latest breakthroughs in quantum computing and summarize the key findings.",
        }
    ],
    tools=[{"type": "web_search"}, {"type": "x_search"}],
)

print(response.output_text or response.model_dump_json(indent=2))

JavaScript Code Example

import OpenAI from "openai";

// Get your CometAPI key from https://api.cometapi.com/console/token, and paste it here
const apiKey = process.env.COMETAPI_KEY || "<YOUR_COMETAPI_KEY>";
const baseUrl = "https://api.cometapi.com/v1";

const client = new OpenAI({
  apiKey,
  baseURL: baseUrl,
});

const response = await client.responses.create({
  model: "grok-4.20-multi-agent-beta-0309",
  input: [
    {
      role: "user",
      content: "Research the latest breakthroughs in quantum computing and summarize the key findings.",
    },
  ],
  tools: [{ type: "web_search" }, { type: "x_search" }],
});

console.log(response.output_text ?? JSON.stringify(response.output, null, 2));

إصدارات Grok 4.20

قد تشمل الأسباب وراء احتواء Grok 4.20 على لقطات متعددة عوامل محتملة مثل اختلاف المخرجات بعد التحديثات التي تتطلب اللقطات القديمة لضمان الاتساق، وتوفير فترة انتقالية للمطورين للتكيف والهجرة، وارتباط اللقطات المختلفة بنقاط النهاية العالمية أو الإقليمية لتحسين تجربة المستخدم. للاطلاع على الاختلافات التفصيلية بين الإصدارات، يرجى الرجوع إلى الوثائق الرسمية.

Model iddescriptionAvailabilityRequest
grok-4.20-multi-agent-beta-0309Multi-agent variant tuned for realtime agent orchestration and tool calling (useful for deep research workflows where multiple sub-agents perform web searches, code execution, and critique).✅response format calls.
grok-4.20-0309-reasoningReasoning-optimized variant: prioritizes deeper chain-of-thought style reasoning and higher benchmark scores on reasoning-heavy tests; higher latency/cost per token expected versus non-reasoning variants.✅chat format call and response format calls.
grok-4.20-0309-non-reasoningLower-latency / lower-cost variant for high-throughput tasks where deterministic, short answers or streaming outputs are the priority; tradeoffs include lower reasoning benchmark scores.✅chat format call and response format calls.