iFlytek launches Spark Token Factory — a platform for intelligent routing and management of enterprise AI models

On July 19, iFlytek officially launched Spark Token Factory, an enterprise-level intelligent routing platform for AI models. The platform addresses challenges enterprises face when scaling AI, such as managing multiple models, controlling costs, ensuring system stability, and achieving unified governance. Spark Token Factory provides unified model access, intelligent routing, full-chain governance, and cost optimization. Its intelligent routing assesses task complexity based on prompt length, rules, dialogue turns, and session affinity, then routes requests to the most suitable model based on quality, cost, latency, availability, and security, with an average decision latency under 100 ms. The inference engine is optimized for Ascend hardware, achieving up to 5x efficiency improvement over the open-source vLLM-Ascend framework in benchmarks, and reducing first-token latency by 30-40%. Security features include separation of duties, audit logs, sensitive data protection, and cost attribution. The platform aims to help enterprises move from single-point AI experiments to large-scale, managed AI operations.