New version of workflow engine goes live with built-in token relay
In October 2026, several leading AI development platforms in China rolled out product updates in rapid succession, with the most notable being a major platform's new version of its workflow engine. For the first time, the engine features a built-in AI relay station function, allowing enterprises to seamlessly switch between different model providers within a single workflow and automatically manage token consumption during inference. According to the platform's official introduction, the core of this feature is an intelligent routing system that dynamically selects the most suitable model endpoint based on task type, latency requirements, and cost budget, while unifying the accounting of token consumption across different models to form a centralized control mechanism similar to a token relay station. This design eliminates the need for enterprises to manually adapt multiple APIs or worry about data format changes when switching models, significantly lowering the barrier to multi-model collaborative deployment.
The platform also announced that the new version has deeply optimized workflow memory and compute scheduling, supporting parallel branches and conditional logic to meet the needs of AI agents in complex business scenarios. Industry analysts believe that the integration of workflow engines and token relay stations marks a critical step in the evolution of AI application development from single-model calls to multi-model orchestration.
New ideas for cost optimization: Dynamic routing and token wholesale
As large model API prices continue to diverge, enterprises are paying more attention to controlling inference costs. Alongside the workflow engine release, the aforementioned platform also introduced a "cost optimization" mode, which uses dynamic routing to direct requests to the cheapest model endpoint that still meets performance requirements, achieving a batch discount effect similar to token wholesale. The platform claims that this mode can reduce the average cost per million tokens by over 30% in real-world testing, and in some scenarios it even achieves what the industry calls the "cheapest tokens." However, it should be noted that "cheapest" here refers to the real-time best price found by the routing algorithm while ensuring response quality, not a fixed low price.
Another well-known AI infrastructure service provider also updated its AI API relay station product this month, adding "model health monitoring" and "automatic failover" features. When a model service experiences an abnormality, the system automatically redirects requests to a backup model and records the additional token cost in a dedicated account, preventing business interruptions due to single points of failure. Such features further enhance the ability to manage token pools, enabling enterprises to manage inference budgets like they manage water and electricity resources.
AI channel proxy model accelerates enterprise adoption
In addition to technical updates, October also saw the emergence of more AI channel proxy partnership models targeting enterprises. Multiple cloud service providers and model vendors have launched "one-stop model access" services, allowing enterprises to call multiple models through an AI large model proxy platform without separate contracts or integrations. This model is especially popular among small and medium-sized enterprises — they don't need to build a dedicated technical team; they simply configure API keys and usage quotas on a unified management interface like an AI hub to quickly deploy AI features.
For example, a token proxy service provider's latest version supports enterprises in setting custom token budget caps and generating detailed cost reports for financial auditing. At the same time, the proxy platform includes built-in compliance filtering and content security modules to ensure model outputs meet regulatory requirements. Industry observers note that the channel proxy model is reshaping the AI supply chain: model vendors focus on R&D, while proxy platforms handle integration, routing, billing, and security, helping to lower the threshold for scaling AI applications.
Outlook: AI agent development accelerates toward platformization
Looking at the trends in October, AI agent development is evolving from scattered tool combinations to unified platform-based solutions. The deep integration of workflow engines, token relay stations, and channel proxy features enables enterprises to build AI applications just like building traditional software — defining data flows, configuring model nodes, and monitoring running costs. In the future, as more standard interfaces and open-source ecosystems mature, the deployment cycle for AI agents could shrink from weeks to hours. However, the industry also faces challenges such as model pricing volatility, routing algorithm fairness, and data sovereignty, all of which require joint exploration by platforms and the entire value chain. Overall, this wave of product updates in October has laid a solid foundation for the upcoming large-scale deployment of AI agent applications.