Vercel Token Pass is a $30-per-month add-on for Pro teams that gives $80 of monthly AI Gateway credit for featured open-weight models. On November 9, 2026, Vercel added GLM 5.3 Flash, and the current docs now let any AI Gateway credential spend the pass and fall back to regular credits.
Last updated: November 9, 2026
Table of contents
- What is new in Token Pass
- Featured models and list prices
- Launch rules compared with the current rules
- How to set up Token Pass in a coding agent
- Routing, fallback and keys
- Limitations to plan for
- Frequently asked questions
- Sources
What is new in Token Pass
Vercel opened Token Pass in early access for Pro teams on October 15, 2026, according to its launch changelog post. The launch covered three models: DeepSeek V4.1 Flash, GLM 5.3 and Kimi K3.
On November 9, 2026, Vercel’s changelog announced that GLM 5.3 Flash had joined the pass. Vercel describes it as Z.ai’s native multimodal coding model with a 1M-token context window. You add it from the Token Pass page in your team’s AI Gateway dashboard.
The bigger change for power users is in the documentation. The Token Pass overview now says any AI Gateway credential spends the pass, and standard credentials fall back to regular AI Gateway credits when the pass runs out. The launch post said requests stopped until the next billing period, with no automatic overage charges.
Token Pass is a pricing layer on top of AI Gateway, so it competes with routing services such as the ones in our OpenRouter Auto Code guide. The difference is the flat monthly fee in exchange for discounted credit on a short list of models.
Featured models and list prices
AI Gateway deducts each request from your credit at the model’s list price, which Vercel defines as the model provider’s own API rate. These are the rates in the Token Pass documentation, in US dollars per million tokens:
| Model | Model ID | Input | Cached input | Output |
|---|---|---|---|---|
| Kimi K3 | moonshotai/kimi-k3 |
$3.00 | $0.30 | $15.00 |
| GLM 5.3 | zai/glm-5.3 |
$1.40 | $0.26 | $4.40 |
| GLM 5.3 Flash (new) | zai/glm-5.3-flash |
$0.15 | $0.03 | $0.50 |
| DeepSeek V4.1 Flash | deepseek/deepseek-v4.1-flash |
$0.15 | $0.003 | $0.60 |
The DeepSeek ID routes to the latest V4.1 Flash weights, currently the 10/17 snapshot. Our DeepSeek V4.1 Flash guide covers that update, and the Kimi K3.1 Pro guide covers Moonshot’s newer model, which is not in the pass. Requests show Vercel Token Pass (token-pass) as the provider, currently served by Wafer.
The cheap Flash models stretch the credit furthest. Because the pass bills at list price, the value comes from paying $30 for $80 of credit, not from lower per-token rates.
Launch rules compared with the current rules
| Area | At launch (October 15) | Current docs (November 9) |
|---|---|---|
| Featured models | DeepSeek V4.1 Flash, GLM 5.3, Kimi K3 | Adds GLM 5.3 Flash |
| Credentials | Launch post says to follow the setup instructions; key types not described | Any AI Gateway credential: standard key, OIDC token or Vercel sign-in |
| Credit runs out | Requests stop until the next billing period | Standard credentials fall back to AI Gateway credits; scoped keys still stop |
| Provider errors | Not described in the launch post | Standard credentials fail over to other providers, billed to AI Gateway credits |
| Price and credit | $30 per month for $80 of credit | Unchanged |
How to set up Token Pass in a coding agent
The getting started guide and the coding agents page describe this sequence:
- Open the Token Pass tab in AI Gateway on a Pro team and click Subscribe, then Enable and Pay. You need permission to manage billing. If Early Access is full, you see a waitlist instead.
- Run
npx vercel@latest ai-gateway setup. It detects installed agents, backs up every file it changes and creates a standard key if you don’t have one. Add--dry-runto preview or--agent <name>for one agent. - Pick a featured model ID in the agent, for example
claude --model zai/glm-5.3orcodex --model zai/glm-5.3. - Turn off any fast mode. Fast variants are not featured models and bill AI Gateway credits instead.
- Point background slots at a featured model too. In Claude Code, set
ANTHROPIC_DEFAULT_HAIKU_MODEL,CLAUDE_CODE_SUBAGENT_MODELand the other model variables in~/.claude/settings.json, or subagent calls bill AI Gateway credits. - Confirm the provider was
token-passin Observability, inproviderMetadata.gateway.routing.resolvedProvider, or withfx ask --json.
Agent-specific notes matter. In fx, picking Kimi K3 preselects Fast, so choose standard speed. In OpenCode, set "model": "vercel/zai/glm-5.3" and remove any options.order that leaves out token-pass; our OpenCode 2.2 guide covers that agent’s config. Vercel also names Pi, Cline and Command Code as compatible agents; see our Pi coding agent 1.5 guide.
Routing, fallback and keys
The models and routing page says a request bills to Token Pass when the team has an active pass with credit, the request uses a featured model ID, and the token-pass provider serves it. With a standard credential, Token Pass is moved to the front of the normal provider chain, so you don’t need providerOptions.
A Token Pass-scoped key is optional. You create it by turning on Restrict to Token Pass only under Advanced Configuration, or with vercel ai-gateway api-keys create --token-pass. It never falls back or fails over, returns a 403 for other models, and its type can’t be changed later. Use it when a hard spending cap matters more than uptime.
All Token Pass errors return HTTP 403 with type no_providers_available. The message tells you whether credit is used up, the key called a non-featured model, or the team has no active pass.
Limitations to plan for
- Pro only. Hobby teams must upgrade, Pro trial teams must move to paid Pro, and Enterprise teams can’t subscribe.
- Shared, expiring credit. The $80 is per team, the first period is prorated, and unused credit doesn’t roll over, according to the billing page.
- Allowlists. If your team uses a provider allowlist, add
token-pass; with a model allowlist, add the featured IDs. Otherwise requests fail. - Early Access terms. Vercel may change the featured models or providers at any time, and may end the pass with a pro-rata refund.
- Fallback costs. After the pass runs out, standard keys keep working only if the team has enough AI Gateway credits, so consider auto top-up or a budget.
For Z.ai’s other recent variant, see our GLM-5.3-FlashX-Legal guide.
Frequently asked questions
What is Vercel Token Pass?
It is a monthly add-on for Vercel Pro teams. For $30 per team per month, the team gets $80 of credit for featured open-weight models served through AI Gateway, with zero data retention. It is in Early Access, and Enterprise teams can’t subscribe.
Which models does Token Pass cover?
As of November 9, 2026, the featured models are Kimi K3 (moonshotai/kimi-k3), GLM 5.3 (zai/glm-5.3), GLM 5.3 Flash (zai/glm-5.3-flash) and DeepSeek V4.1 Flash (deepseek/deepseek-v4.1-flash). Fast variants, image, video, audio and embedding models are not covered.
What happens when Token Pass credit runs out?
With a standard AI Gateway API key or an OIDC token, featured-model requests fall back to your team’s AI Gateway credits at the serving provider’s rates. With a Token Pass-scoped key, requests fail with a 403 error until the next billing period.
Do I need a special API key for Token Pass?
No. Vercel’s current documentation says any AI Gateway credential works, including a standard API key, an OIDC token or signing in with your Vercel account. A Token Pass-scoped key is optional and caps spending at the pass.
Does unused Token Pass credit roll over?
No. Credit expires at the end of each billing period and resets to $80 when the next period starts. The credit is shared by everyone on the team, and the first period is prorated.
Sources
- Vercel: Token Pass documentation
- Vercel: Token Pass models and routing
- Vercel: Using Token Pass with coding and personal agents
- Vercel changelog: Introducing Token Pass (October 15, 2026)
- Vercel changelog: GLM 5.3 Flash is now available in Token Pass (November 9, 2026)
Image: Vercel announcement graphic from the Vercel changelog, used for editorial purposes. FreeAnswers is not affiliated with Vercel.