---
title: "auto/security: an LLM router for security work"
canonical: https://pentestllm.ai/auto-router
generated: 2026-11-11T18:04:26.129Z
source: https://pentestllm.ai/auto-router
base_url: https://api.pentestllm.ai/v1
---

## How the router decides

The router reads what the request carries, on our side and with no extra model call, and sends it to one of three tiers. Each tier lists its models in the order the router tries them; prices are from the live catalog as of 2026-11-11. The answer’s `model` field names the build, the `x-uplink-auto-tier` header names the tier, and `auto.tier_max` caps how high a request may go.

Fast

Short questions and plain text: everything the other two tiers do not claim.

- Gemma 4 26B A4B Cybersecurity: $0.4224 in, $1.3376 out per million tokens
- GLM-5.3 Flash Cybersecurity: $0.5093 in, $1.6975 out per million tokens
- Qwen3.8 27B Cybersecurity: $0.40 in, $2.40 out per million tokens

Balanced

Tools, a code block, an image, a json_schema answer, or a prompt over 16,000 tokens.

- Red Team (Flash): $0.90 in, $3.00 out per million tokens
- OrcaCyber MiniMax M3.1 Flash: $2.00 in, $4.00 out per million tokens
- OrcaCyber Qwen3.8 Flash Next: $2.00 in, $4.00 out per million tokens

Frontier

A request that asks for careful reasoning or sets a high reasoning effort, or tools together with code.

- Red Team (Pro): $1.50 in, $6.00 out per million tokens
- Cyber Frontier: $5.25 in, $15.75 out per million tokens
- OrcaCyber Zero 1.5: $7.50 in, $18.75 out per million tokens

What the router changes

- Model sent: auto/security or any name, on a router key
- Price: the chosen model’s no router fee
- Session: 30 minutes one model per x-session-id
- Answer: model field + x-uplink-auto-tier which build and tier answered

## A router key for agents with a fixed model name

Some agents pin a model name you cannot edit. A router key fixes that at the key: on [Create key](https://app.pentestllm.ai/console/keys), set Auto router to `auto/security`, and every chat request that key sends is routed as that id whatever model the client names. Existing keys have the same setting.

## Requests

A tier cap and a sticky session in curl, then the OpenAI SDK in Python on the same base URL.

curl

```
curl https://api.pentestllm.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-ptl-live-…" \
  -H "Content-Type: application/json" \
  -H "x-session-id: engagement-42" \
  -d '{
    "model": "auto/security",
    "auto": {"tier_max": "balanced"},
    "messages": [{"role": "user", "content": "Write a Sigma rule for repeated failed SSH logins from one source."}]
  }'
```

python

```
from openai import OpenAI

client = OpenAI(api_key="sk-ptl-live-…", base_url="https://api.pentestllm.ai/v1")

reply = client.chat.completions.create(
    model="auto/security",
    messages=[{"role": "user", "content": "Explain the CVSS 4.0 base metrics in five lines."}],
    extra_headers={"x-session-id": "engagement-42"},
)
print(reply.model)  # the build that answered
print(reply.choices[0].message.content)
```

## Questions

### What does model "auto/security" pick?

A build carrying the security tag in the catalog, chosen on our side for each request with no extra model call. A reasoning ask or tools together with code go to the frontier tier; tools, a code block, an image, a json_schema answer or a prompt over 16,000 tokens go to balanced; everything else goes to fast. The answer’s model field names the build and the x-uplink-auto-tier header names the tier.

### What does the router cost?

The chosen model’s own price per million tokens, as listed on its model page. There is no router fee and no markup for sending the router id instead of a model id.

### Does an agent run keep the same model?

Yes, with an x-session-id header: the model that answered a session keeps answering it for 30 minutes from its last request, so a run does not change model between steps.

### Can I cap the tier?

Yes. "auto": {"tier_max": "balanced"} keeps a request out of the frontier tier; "fast" keeps it in the fast tier.

### Which endpoints take the router id?

Chat completions, Messages and Responses, on the same base URL and key. "auto/security" also answers to "auto/cyber".

### What is an auto router key?

A key with the Auto router setting on. Every chat request it sends is routed as auto/security whatever model name the client sends, which helps with an agent that pins a model name you cannot edit. Set it when you create a key, or in an existing key’s settings in the console.

### Will the router answer anything?

No. The security builds refuse less often than stock models on security questions, not never, and the acceptable use policy applies to every request: authorized testing only, on systems you own or have written permission to test.

## Read next

- [Every security and abliterated build, priced](https://pentestllm.ai/models)
- [API docs: base URL, key, server tools](https://pentestllm.ai/docs)
- [Agent setup](https://pentestllm.ai/agents)
- [Acceptable use: authorized testing only](https://pentestllm.ai/acceptable-use)