At today's Baidu World 2026 conference, Baidu founder Robin Li officially unveiled a brand-new large model service platform — “Token Exchange”. Addressing the current pain points of “choice paralysis” and “lack of trust” that enterprises and developers commonly face when dealing with massive numbers of large models, the platform innovatively introduces the operational concepts of a stock exchange, aiming to provide the industry with a neutral and trusted venue for model trading and distribution.
Analogous to a stock exchange: gathering high-quality resources and model providers
In daily development and commercial deployment, enterprise users often find that large models on the market are numerous in variety, vary in price, come from uneven sources, and their actual performance is often hard to predict.
The positioning of “Token Exchange” is precisely to solve this pain point. Like a stock exchange for the large model industry, on one hand it attracts and gathers quality models and model providers, and on the other hand it builds a convenient, trusted procurement and invocation channel for enterprises and individual developers, allowing users to select and use model resources suited to their own business with greater confidence.
Strict entry control and continuous monitoring: firmly resisting inflated parameters and model substitution
To ensure the quality and reliability of model resources on the platform, “Token Exchange” has built a strict screening and monitoring mechanism.
When a resource is first listed, the platform conducts comprehensive entry testing, covering multiple stages including the model itself, fingerprint comparison, model capability testing, performance metrics testing, and availability testing. On this basis, the platform also conducts continuous monitoring of model quality parameters, using technical means to effectively prevent industry problems such as “model substitution” or “inflated parameters”, ensuring that every model resource has tangible quality guarantees.
Establishing public credit records: multi-dimensional scoring and dynamic risk alerts
In addition to upfront entry testing, “Token Exchange” has also established a transparent credit evaluation system.
The platform comprehensively scores model providers across multiple dimensions including service quality, fulfillment, security compliance, and commercial contract performance, and grants corresponding credit rating labels. Once certain model providers' credit declines, the platform promptly issues proactive alerts to users, and all evaluation and scoring data ultimately accumulates into publicly searchable credit records.
According to data disclosed at the conference, within just one month of launch, the number of resources supplied by “Token Exchange” has exceeded 100, and the daily average number of Tokens invoked has surpassed the 70 billion mark, demonstrating strong market growth momentum.