A.X K1 Technical Report

TL;DR

A.X K1 is a large Mixture-of-Experts language model trained on 10 trillion tokens, designed for scalable deployment and controllable reasoning, with competitive performance and a novel Think-Fusion training method.

Contribution

Introduces A.X K1, a 519B-parameter MoE model with a new training recipe for user-controlled reasoning modes and optimized for diverse real-world applications.

Findings

Achieves competitive performance with leading models

Demonstrates superior results on Korean-language benchmarks

Supports explicit controllable reasoning modes

Abstract

We introduce A.X K1, a 519B-parameter Mixture-of-Experts (MoE) language model trained from scratch. Our design leverages scaling laws to optimize training configurations and vocabulary size under fixed computational budgets. A.X K1 is pre-trained on a corpus of approximately 10T tokens, curated by a multi-stage data processing pipeline. Designed to bridge the gap between reasoning capability and inference efficiency, A.X K1 supports explicitly controllable reasoning to facilitate scalable deployment across diverse real-world scenarios. We propose a simple yet effective Think-Fusion training recipe, enabling user-controlled switching between thinking and non-thinking modes within a single unified model. Extensive evaluations demonstrate that A.X K1 achieves performance competitive with leading open-source models, while establishing a distinctive advantage in Korean-language benchmarks.