> ## Documentation Index > Fetch the complete documentation index at: https://benchgen.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Inference API Token > Issue a new inference token from a model's page for direct API usage. Every deployed model has its own **API Keys** tab where you generate a key scoped to that model, then call its endpoint directly, from `curl`, your own backend, or any OpenAI-compatible client. This works on any deployed model, a router like `benchgen-router-lite`, a model you uploaded, or one you fine-tuned in Train. If you don't have a deployed model yet, see [Deploy an Inference Model](/docs/eval/run-an-inference-model). *** ## Step 1: Open your model In **Models**, find the model you want a key for, it needs to be **Deployed** or **Running** to serve an endpoint. Here, [`benchgen-router-lite`](https://benchgen.com/models/benchgen/benchgen-router-lite) is running. The AI Models page listing benchgen-router-lite under Running, and deployed models such as CRM Agent and Finance Agent below it

The AI Models page listing benchgen-router-lite under Running, and deployed models such as CRM Agent and Finance Agent below it

Open it. Its card has four tabs: **Model Card**, **Usage**, **API Keys**, and **Evaluations**. Notice the sidebar on the right already carries an **API key** field with a **Generate key** shortcut, that same control is there no matter which tab you're on. A model card's Model Card tab, showing tab navigation for Model Card, Usage, API Keys, and Evaluations, and an API key field with a Generate key button in the sidebar

A model card's Model Card tab, showing tab navigation for Model Card, Usage, API Keys, and Evaluations, and an API key field with a Generate key button in the sidebar

## Step 2: The quick way, or the full API Keys tab Click that sidebar **Generate key** button any time you just need a key fast. For naming keys or seeing everything you've already issued, click the **API Keys** tab instead, it's the same action dialog either way, just with more around it. The API Keys tab showing the endpoint URL with a Copy button, a Key name input, a Generate key button, and a table of four active keys with their creation date, last used, spend, and a Revoke button

The API Keys tab showing the endpoint URL with a Copy button, a Key name input, a Generate key button, and a table of four active keys with their creation date, last used, spend, and a Revoke button

Optionally name the key (useful if you're issuing separate keys per app or environment), then click **+ Generate key**. ## Step 3: Copy the key now, it won't be shown again A dialog shows your new **secret key** plus a ready-to-run example request. A dialog titled Your new API key, warning that only a hash is stored so the key cannot be recovered later, with the secret key, a Copy button, and an example curl request against the chat completions endpoint

A dialog titled Your new API key, warning that only a hash is stored so the key cannot be recovered later, with the secret key, a Copy button, and an example curl request against the chat completions endpoint

**Copy it now.** BenchGen stores only a hash of the key, so if you close this dialog without saving it, you can't recover it, you'd have to generate a new one. If a key is ever exposed, revoke it and generate a replacement. Click **Copy** to grab just the key, or copy the example request as-is: ```bash theme={null} curl https://gateway.benchgen.com/v1/chat/completions \ -H "Authorization: Bearer sk-..." \ -H "Content-Type: application/json" \ -d '{"model":"benchgen-router-lite","messages":[{"role":"user","content":"hi"}]}' ``` Click **I have saved it** to dismiss the dialog. ## Managing keys afterward Your new key appears in the table on the API Keys tab, **Active**, with its creation date, last-used timestamp, and running spend. Click **Revoke** on any key you no longer want live, it flips to a **Revoked** status immediately and stops authenticating. The same keys table with one key now showing a Revoked badge instead of Active and a Revoke button

The same keys table with one key now showing a Revoked badge instead of Active and a Revoke button

*** ## Watching it get used Every call made with the key rolls up into the model's **Usage** tab: total requests, tokens, spend, and average latency, plus a log of every individual request. The Usage tab showing 2 requests, 175 tokens, $0.001526 spend, 4066 ms average latency, and a Requests table with two successful calls

The Usage tab showing 2 requests, 175 tokens, $0.001526 spend, 4066 ms average latency, and a Requests table with two successful calls

Click any row to see that exact call in full: its tags, request details, token and cost metrics, and the actual prompt and response. A request detail drawer showing owner and source tags, request details including model and provider, token and cost metrics, and the start of the request and response payload

A request detail drawer showing owner and source tags, request details including model and provider, token and cost metrics, and the start of the request and response payload

See [Monitor Model Usage](/docs/eval/monitor-model-usage) for the full breakdown of everything on this tab. *** ## Next Steps Track requests, tokens, spend, and inspect individual calls made with your key. Point your key at the router endpoint instead of a single model. Open the model card these screenshots were taken from.