Documentation

Atlas Core (Models)

The foundation of the MX4 AI platform.

Last updated on February 2, 2026

Atlas Core represents our family of sovereign large language models, trained specifically for high-performance Arabic and English reasoning.

Available Models

mx4-atlas-core

Our flagship model. Best for complex reasoning, content generation, and Arabic dialect understanding.

8k ContextGeneral Purpose

mx4-atlas-lite

Optimized for speed and cost. Ideal for classification, extraction, and real-time chat.

16k ContextLow Latency

Model Specifications

Model	Parameters	Context	Latency	Throughput
mx4-atlas-core	70B	8,192 tokens	150-250ms	150 req/min
mx4-atlas-lite	13B	16,384 tokens	50-100ms	500 req/min

Performance Benchmarks (Arabic Tasks)

Benchmark	Atlas Core	Atlas Lite	GPT-4
Arabic QA (ARCD)	92.3%	88.7%	91.8%
Sentiment Analysis	89.2%	85.6%	87.4%
Named Entity Recognition	94.1%	91.3%	93.2%
Dialect Understanding (5 dialects)	90.8%	87.2%	85.3%

Use Case Recommendations

Choose Atlas Core for:

Complex reasoning and analysis tasks
Long-form content generation
Multi-dialect Arabic conversation
Nuanced translation tasks
Fine-tuning on specialized domains

Choose Atlas Lite for:

Real-time chat applications
Text classification & extraction
High-volume API usage (cost sensitive)
Edge deployment / low-latency requirements
Structured output tasks (JSON, schemas)

Pricing Guide

Atlas Core

70B parameter model

$0.50 / 1M input tokens

$1.50 / 1M output tokens

Atlas Lite

13B parameter model

$0.10 / 1M input tokens

$0.30 / 1M output tokens

Quick Start

quick_start.pypython

1import openai
2import os
3
4client = openai.OpenAI(
5    api_key=os.getenv("MX4_API_KEY"),
6    base_url="https://api.mx4.ai/v1"
7)
8
9# Use Atlas Core for complex tasks
10response = client.chat.completions.create(
11    model="mx4-atlas-core",  # Flagship model
12    messages=[
13        {"role": "user", "content": "اشرح لي مفهوم الذكاء الاصطناعي بطريقة مبسطة"}
14    ],
15    temperature=0.7
16)
17print("Core:", response.choices[0].message.content)
18
19# Use Atlas Lite for fast responses
20response = client.chat.completions.create(
21    model="mx4-atlas-lite",  # Lightweight model
22    messages=[
23        {"role": "user", "content": "صنف المشاعر: هذا المنتج ممتاز!"}
24    ],
25    temperature=0.3
26)
27print("Lite:", response.choices[0].message.content)