skt/A.X-4.0

TEXT GENERATIONPricing:Input $12 / Output $20Concurrent Unit Cost:4Model Size:72.7BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 2, 2025License:otherArchitecture:Transformer0.2K Featherless Exclusive Cold

SK Telecom's A.X 4.0 is a 72.7 billion parameter large language model, built upon the Qwen2.5 architecture and optimized for Korean-language understanding and enterprise deployment. It demonstrates superior Korean proficiency, scoring 78.3 on KMMLU and 83.5 on CLIcK, outperforming GPT-4o in these benchmarks. The model also features efficient token usage for Korean input and supports a long context window of up to 131,072 tokens, making it suitable for processing extensive documents and conversations.

Loading preview...

A.X 4.0 Overview

SK Telecom's A.X 4.0 is a large language model (LLM) specifically optimized for Korean-language understanding and enterprise applications. Released on July 03, 2025, it is built on the open-source Qwen2.5 model and has been extensively trained with large-scale Korean datasets.

Key Capabilities & Differentiators

  • Superior Korean Proficiency: Achieves 78.3 on KMMLU and 83.5 on CLIcK, surpassing GPT-4o in both leading Korean-language benchmarks.
  • Deep Cultural Understanding: Excels in Korean cultural and contextual comprehension.
  • Efficient Token Usage: Utilizes approximately 33% fewer tokens than GPT-4o for the same Korean input, leading to more cost-effective processing.
  • Long Context Handling: Supports an impressive context length of up to 131,072 tokens, enabling the comprehension of lengthy documents and complex conversations.
  • Deployment Flexibility: Available in a 72B-parameter standard version (A.X 4.0) and a 7B lightweight version (A.X 4.0 Light).

Ideal Use Cases

A.X 4.0 is particularly well-suited for applications requiring advanced Korean language processing, deep cultural understanding, and efficient handling of long-form text. Its enterprise optimization makes it a strong candidate for business environments in Korea, including customer service, content generation, and data analysis where Korean linguistic nuance is critical. The model also supports tool-use functionality, enhancing its utility for complex tasks requiring external function calls.