Vercel Token Pass Launches: $30/Month Access to GLM 5.3 and Kimi K3

Vercel has officially launched Token Pass, an AI model subscription service for Pro teams. For $30 per month, teams get $80 worth of monthly credits to call two models: GLM 5.3 and Kimi K3. This is the first attempt in the AI gateway space to bundle Chinese large models with a subscription model, reflecting a shift in competition at the AI infrastructure layer—from "how many models can we connect" to "how can we make it easier for teams to actually use models."

Product Mechanics: Subscription Pricing and Transparent Credits

The core logic of Token Pass is "prepaid credits, consumed on demand." A $30 monthly fee buys $80 in model usage credits—roughly 2.7x leverage. Compared to pay-as-you-go pricing, this model reduces the complexity of cost accounting when teams switch between multiple models.

Once credits are exhausted, requests stop and no automatic overage charges are incurred. Unused credits do not roll over to the next month. This design may seem strict, but it actually prevents teams from receiving unexpectedly high bills due to overspending, making it friendlier for budget-conscious small and mid-sized teams. Subscriptions renew monthly and can be canceled at any time.

Token Pass is currently in early access and only available to Pro teams.

Why GLM 5.3 and Kimi K3

Choosing these two models as the launch lineup is an intriguing signal. GLM 5.3 comes from Z.ai, and Kimi K3 comes from Moonshot AI. Both are standout Chinese models in code generation and reasoning tasks. By bundling them into a developer-facing subscription, Vercel is clearly targeting teams seeking cost-effective models for coding agent scenarios.

The company explicitly states that "more models are coming soon," meaning Token Pass isn't positioned around specific models—it aims to become the entry point for teams to centrally manage model access.

Z.ai and Zhipu's GLM Series: GLM comes from Zhipu AI, whose overseas brand is Z.ai. The series initially gained influence through an open-source approach, and versions like GLM-4 are competitive in Chinese language understanding and tool calling. GLM 5.3 is a newer iteration in the series, focused on code generation and multi-step reasoning, often used in agent workflows for planning, function calling, and code repair. For developers, choosing GLM typically means relatively controllable costs and solid engineering compatibility.

Moonshot AI and the Kimi Series: Kimi was launched by Moonshot AI and first entered public view with its ultra-long context handling capabilities, with later versions gradually strengthening reasoning and programming performance. Kimi K3 continues this direction and is widely used for building coding assistants in long-document understanding, complex instruction following, and code tasks. Bundling both models into a single subscription essentially packages "the cost-performance advantage of Chinese models in coding agent scenarios" directly for overseas developer teams.

Zero Data Retention Enabled by Default

Token Pass has Zero Data Retention (ZDR) enabled by default. Requests run through the selected providers, with inference occurring in the US and EU. After each request completes, the AI Gateway and providers delete the prompt and response content, and user prompts are not used to train models.

For enterprise teams, data security is often the primary concern when adopting third-party AI services. Having ZDR enabled by default rather than as an optional feature reduces friction in compliance reviews.

Zero Data Retention (ZDR) is a data governance commitment: after processing a request, the service provider does not persistently store prompts and model outputs, nor use them for model training or human review. It differs from weaker commitments like "opt out of training" and typically requires providers to make targeted changes to transmission pipelines, inference runtimes, and logging systems—which is why it's more common in enterprise-facing paid products. For teams that need to satisfy SOC 2, GDPR, or industry compliance reviews, ZDR can significantly simplify the evaluation process for third-party AI services. Note that ZDR generally covers content retention during inference, but metadata such as billing and rate limits is usually still recorded.

Integration with Coding Agents

Token Pass can be used directly with coding agent tools like fx, Command Code, and Pi, or you can build custom personal agents through the AI SDK. All requests are routed through the AI Gateway, and teams can switch between included models while sharing the same credit pool.

This design gives model selection back to developers while retaining unified billing and permission management. For teams already using the Vercel ecosystem, Token Pass is a low-friction incremental option.

AI Gateway and AI SDK: The AI Gateway is a middleware layer between applications and model providers that centrally handles authentication, routing, retries, billing, and observability—so teams don't need to maintain separate keys and quotas for each model vendor. Vercel's AI Gateway is the underlying layer powering Token Pass. The AI SDK is an open-source library for developers that provides a unified model calling interface along with primitives like streaming output and tool calling, allowing developers to switch backend models with the same codebase. Combined, teams can quickly integrate using existing agent tools or build their own agents while reusing the same credits and permission system.

Getting Started

The onboarding process isn't complicated: open the Token Pass page in your team's AI Gateway dashboard, activate the subscription and create an API key, connect your agent following the instructions, then select GLM 5.3 or Kimi K3 to get started.

Token Pass is essentially an experiment in "AI model subscription-ization." Whether developers embrace it depends on how quickly the model lineup expands and how well subscription credits match actual usage. The $30-for-$80 pricing looks attractive on paper, but teams consume model credits at different paces, so whether it's truly worth it will need to be validated through actual use.

One-line Summary

Vercel launches Token Pass: $30/month gives Pro teams $80 in credits for GLM 5.3 and Kimi K3.

Paragraph Summary

Vercel has officially released Token Pass, an AI model subscription service for Pro teams: $30 per month for $80 worth of model usage credits, valid for GLM 5.3 and Kimi K3. The launch lineup features two Chinese models, targeting cost-effectiveness needs in coding agent scenarios. The product has Zero Data Retention (ZDR) enabled by default, with inference in the US and EU, and prompts and responses deleted after requests complete. Once credits run out, requests stop with no automatic overage; unused credits don't roll over and subscriptions can be canceled monthly. Token Pass integrates directly with coding agents like fx, Command Code, and Pi, or you can build custom agents via the AI SDK—all requests route through the AI Gateway and share credits. It's currently in early access, with more models promised soon; its success depends on lineup expansion speed and how well credits match actual usage.