iFlytek Releases Spark X2-Flash Model, Plans to Launch Flagship Large Model on Ascend 950 Platform in October

iFlytek officially released the Spark X2-Flash model today, with API interfaces opened simultaneously. The model uses a MoE architecture, has a total of 30B parameters, supports a maximum context length of 256K, and has significantly improved capabilities in agents and code. It was trained on Huawei's Ascend 910B cluster.

Wu Xiaoru, president of iFlytek, stated at the performance briefing that the company will release China's first flagship large model on the Ascend 950 platform in October this year, which benchmarks the industry's most advanced mainstream models. In addition, the company's Q1 2026 operating revenue was 5.274 billion yuan, a year-on-year increase of 13.23%; net profit attributable to shareholders was -170 million yuan, a year-on-year increase of 12.17%.