---
title: "LLM API Cost Calculator for AI Pentests"
canonical: https://pentestllm.ai/tools/engagement-cost
generated: 2026-11-11T18:05:30.240Z
source: https://pentestllm.ai/tools/engagement-cost
base_url: https://api.pentestllm.ai/v1
---

- Engagement, 10 days of Cyber Frontier: $124.95
- Per day: $12.50
- Per request: $0.0312

| Credit | Requests | Days |
| --- | --- | --- |
| $10 | 320 requests | 0.8 days |
| $50 | 1,600 requests | 4 days |
| $200 | 6,402 requests | 16 days |

## What drives the API cost

A model API bills by the token, with one price for input and a higher one for output. Input is everything a request carries: the system prompt, the conversation so far, tool definitions and whatever tool or scan output you pass back. Output is what the model writes. An agent resends its context on every turn, so in agent-driven security work input tokens usually dominate the bill, and the number of requests matters more than the length of any one answer.

## How to read the estimate

The estimator multiplies your average request by the model’s prices, then by requests per day and by days. Cost per request tells you which model fits the task. Cost per day tells you whether the pace is sustainable. The engagement figure is the number to put in a proposal, with headroom: these are averages you typed in, not a quote, and the real bill follows the tokens actually used. The last rows show how many requests and days of that usage $10, $50 and $200 of prepaid credit cover.

## Ways to bring the cost down

- Keep the stable part of the prompt first and unchanged, so a model with a cached-input price can bill it at the cached rate.
- Trim tool output before passing it back. A summary of a long scan result costs a fraction of the raw text on every later turn.
- Send routine steps to a cheaper build and keep the larger model for the hard reasoning.

## Prices used on this page

Read from the live catalog on 2026-11-11, in US dollars per million tokens. The [pricing page](https://pentestllm.ai/pricing) has the full list and how prepaid credit works.

| Model | Input | Cached input | Output |
| --- | --- | --- | --- |
| Cyber Frontier | $5.25 | $0.9625 | $15.75 |
| Gemma 4 26B A4B Cybersecurity | $0.4224 | none | $1.3376 |
| GLM-5.3 Flash Cybersecurity | $0.5093 | $0.2546 | $1.6975 |
| OrcaCyber MiniMax M3.1 Flash | $2.00 | $0.375 | $4.00 |
| OrcaCyber MiniMax M3.1 Flash | $2.00 | $0.375 | $4.00 |
| OrcaCyber Qwen3.8 Flash Next | $2.00 | $0.375 | $4.00 |
| OrcaCyber Qwen3.8 Flash Next | $2.00 | $0.375 | $4.00 |
| OrcaCyber Zero 1.5 | $7.50 | $1.875 | $18.75 |
| Qwen3.8 27B Cybersecurity | $0.40 | none | $2.40 |
| Red Team (Flash) | $0.90 | $0.1543 | $3.00 |
| Red Team (Pro) | $1.50 | $0.425 | $6.00 |
| Abliterated Large Flash | $2.50 | none | $5.00 |
| Abliterated Model | $2.50 | none | $7.50 |
| Abliterated Model Large | $7.50 | none | $12.50 |
| Abliterated Model Large V2 | $7.50 | $0.75 | $12.50 |
| Apex Flash 1 Abliterated | $0.68 | $0.36 | $2.32 |
| DeepSeek R1 Distill Llama 70B Abliterated | $2.80 | none | $2.80 |
| DeepSeek R1 Distill Qwen 32B Abliterated | $5.60 | none | $5.60 |
| DeepSeek V4 Pro Abliterated | $1.00 | $0.10 | $4.00 |
| DeepSeek V4 Pro Qwen3.5 9B Uncensored Heretic | $1.12 | $0.56 | $1.12 |
| DeepSeek V4.1 Flash 1017 Abliterated | $0.75 | $0.1375 | $3.125 |
| DeepSeek V4.1 Flash 1017 Heretic | $0.825 | none | $3.30 |
| DeepSeek V4.1 Flash 1017 Heretic Thinking | $0.825 | none | $3.30 |
| DeepSeek V4.1 Flash Abliterated | $0.8488 | $0.1455 | $1.6975 |
| DeepSeek V4.1 Flash Abliterated Fast | $1.75 | $0.40 | $3.50 |
| Gemma 31B Heretic Uncensored | $1.50 | none | $8.25 |
| Gemma 4 31B Cognitive Unshackled | $1.071 | none | $1.071 |
| GLM-4.6 Derestricted v5 | $1.40 | none | $5.25 |
| GLM-5.3 Abliterated | $2.1825 | $0.6402 | $4.365 |
| GLM-5.3 Flash Abliterated | $0.4656 | $0.1552 | $1.94 |
| GLM-5.3 Flash Apex Abliterated | $0.80 | $0.32 | $3.20 |
| Llama 3 70B Abliterated v3.5 | $2.80 | none | $2.80 |
| Llama 3.3 70B Abliterated | $2.80 | none | $2.80 |
| MiMo V2.6 Flash Abliterated | $2.00 | $1.10 | $6.00 |
| NeuralDaredevil 8B Abliterated | $1.76 | none | $1.76 |
| Qwen 2.5 32B Abliterated | $1.75 | none | $1.75 |
| Qwen3.5 27B Blossom V6.4 Derestricted | $1.071 | none | $1.071 |
| Qwen3.5 27B BlueStar v3 Derestricted | $1.071 | none | $1.071 |
| Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted | $1.071 | none | $1.071 |
| Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted Lite | $1.071 | none | $1.071 |
| Qwen3.5 27B Queen Derestricted | $1.071 | none | $1.071 |
| Qwen3.8 27B Obliterated | $0.582 | none | $7.27 |

## Questions

### How many tokens does a pentest agent use per request?

It depends on how much context the agent resends each turn. A token is roughly four characters of English, so a few pages of tool output and conversation history is several thousand input tokens, while the answer is usually a few hundred. Run the agent for a day, read the real averages from your usage log, and put those in.

### Are these the prices I would pay?

Yes. The estimator reads each model’s price per million tokens from the live catalog, the same figures the pricing page lists, and the date they were read is shown under the model. Credit is prepaid, and the pricing page explains how it works.

### Does the estimate include scan and tool output?

Only if you send that output to the model. Whatever goes into a request counts as input tokens: the system prompt, the conversation so far, tool definitions and any tool output you pass back. Raise the input figure to match.

### What is cached input?

Some models bill input the model has already read in a recent request, such as an unchanged system prompt, at a lower cached price. Where the catalog lists a cached price, the estimator lets you set the share of input that is served from cache. Where it does not, all input is billed at the input price.

### Is the usage I enter sent anywhere?

No. The prices arrive with the page and the arithmetic runs in your browser. Nothing you type is sent to a server or stored.

## More tools, prices and docs