At the opening of Baidu World 2026 on November 12, Baidu formally launched its large-model trading platform, the Token Exchange, building a trusted marketplace for LLM tokens. According to NetEase Tech, within one month of going live the platform has hosted more than 100 model supply resources, handled over 50 million calls a day and moved more than 700 billion tokens daily.
The platform runs on two mechanisms: verification and credit. Verification checks the authenticity of listed models through fingerprint comparison, capability testing, and performance and availability tests. Credit ratings for providers are built from regular sampling probes and full-chain, real-time monitoring of calls, giving buyers a track record to consult before choosing a model.
Baidu founder Robin Li traced the business back to the pace of the industry: in the first nine months of this year, more than one new model has appeared almost every day, and call demand has grown just as fast. That boom produced a crowd of token resellers that take no responsibility for model quality, service stability or response speed, and whose listings cannot even be verified as genuine. The Token Exchange is aimed squarely at that trust gap.
Trading tokens as a standardized commodity means suppliers must stand behind continuous availability checks, while buyers can compare options against public ratings. For Baidu, the marketplace also moves its own cloud's model distribution onto more neutral ground: one side connects models and providers, the other serves enterprises and individual developers, with revenue tied directly to call volume.
For the industry, the exchange turns the open question of "which model to pick" into a shelved choice backed by credit ratings. How much those ratings are worth will depend on how strictly the probes are run and how often scores are refreshed. That is the market test the platform now faces.