What a BYOK AI writing assistant actually costs
Bring your own key means you pay the AI provider per use instead of a monthly subscription. Here are the numbers for everyday writing.
Published
The short answer
With your own API key, an AI-written email reply costs roughly $0.0002 to $0.03, depending on the model. At 20 replies every working day, that is about $0.08 to $0.80 a month on small and mid-size models (DeepSeek V4 Flash, GPT-5 nano, GPT-4o mini, Gemini 2.5 Flash), about $6.60 on Claude Sonnet 4.6, and about $12 on GPT-5.5. Writing subscriptions cost $10 to $12 a month. Free models and local models bring the cost to zero.
Prices are OpenRouter list prices on 26 September 2026, in US dollars per million tokens.
Cost per request and per month
| Model | Input / 1M tokens | Output / 1M tokens | Short rewrite | Reply with page context | Answer from a PDF | 440 replies a month |
|---|---|---|---|---|---|---|
| DeepSeek V4 Flash | $0.05 | $0.09 | $0.00005 | $0.0002 | $0.0006 | $0.08 |
| OpenAI GPT-5 nano | $0.05 | $0.40 | $0.0001 | $0.0003 | $0.0007 | $0.13 |
| Google Gemini 2.5 Flash-Lite | $0.10 | $0.40 | $0.0002 | $0.0005 | $0.0013 | $0.21 |
| OpenAI GPT-4o mini | $0.15 | $0.60 | $0.0002 | $0.0007 | $0.0019 | $0.31 |
| Google Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $0.0005 | $0.0013 | $0.0034 | $0.58 |
| Google Gemini 2.5 Flash | $0.30 | $2.50 | $0.0007 | $0.0018 | $0.0043 | $0.79 |
| Anthropic Claude Haiku 4.5 | $1.00 | $5.00 | $0.0017 | $0.0050 | $0.013 | $2.20 |
| Anthropic Claude Sonnet 4.6 | $3.00 | $15.00 | $0.0051 | $0.015 | $0.039 | $6.60 |
| OpenAI GPT-5.5 | $5.00 | $30.00 | $0.0095 | $0.027 | $0.067 | $11.66 |
How we estimated
A token is about four characters of English text. Each request has three parts: instructions, context and the answer.
- Short rewrite: 700 input tokens, 200 output. A paragraph you want fixed or reworded, plus the tool's instructions (about 500 tokens in ShroomPen).
- Reply with page context: 3,500 input tokens, 300 output. The email thread or page you are replying to. ShroomPen caps page context at 12,000 characters, about 3,000 tokens.
- Answer from a PDF: 11,000 input tokens, 400 output. ShroomPen caps a PDF at 40,000 characters, about 10,000 tokens.
- Per month: 20 replies a day for 22 working days.
Compared with subscriptions
| Option | Price |
|---|---|
| Grammarly Pro | From $12 a month, 2,000 AI prompts a month (grammarly.com/plans) |
| Wandpen Pro | $10 a month, or $8.30 a month billed yearly |
| Wandpen BYOK license | $19 once (listed as limited time, normally $49), plus your provider costs |
| ShroomPen | Free, plus your provider costs |
A subscription buys you predictability and no setup. BYOK buys you a lower bill for normal use, the choice of model, and a direct relationship with the provider. If you write hundreds of long answers a day with the most expensive models, a flat subscription can work out cheaper. The Wandpen comparison goes into the trade-offs.
Ways to pay nothing
- OpenRouter free models. 50 requests a day, or 1,000 after you buy $10 of credits once. See which free models suit writing.
- Gemini API free tier. Gemini 2.5 Flash and Flash-Lite are free within Google's rate limits. Google says content on the free tier is used to improve its products, so keep private text off it.
- A local model. Ollama or LM Studio on your own computer costs only electricity, and the text stays on your machine.
Extra costs to know about
- OpenRouter adds a 5.5% fee ($0.80 minimum) when you buy credits by card. It does not mark up model prices.
- Paid models usually need prepaid credit or a card on file with the provider.
- Set a monthly spending limit in your provider's dashboard so a mistake can't run up a bill.
Try it with your own key
Install ShroomPen, press Ctrl+Q on any page and connect OpenRouter, OpenAI, Anthropic, Gemini or a local server. There is no ShroomPen fee.
Related pages
Questions
How much does one AI email reply cost with my own API key?
With a reply that reads about a page of context, between about $0.0002 and $0.03 in our estimate, depending on the model. Small models such as GPT-5 nano, DeepSeek V4 Flash or GPT-4o mini cost well under a tenth of a cent. Claude Sonnet 4.6 costs about 1.5 cents.
Is BYOK cheaper than a subscription?
For most people, yes. 440 replies a month (20 per working day) costs about $0.10 to $0.80 on small and mid-size models, against $10 to $12 a month for Wandpen Pro or Grammarly Pro. Heavy use of a top model such as GPT-5.5 can reach the price of a subscription.
Can it cost nothing at all?
Yes, three ways: OpenRouter's free models (50 requests a day, or 1,000 after a one-time $10 top-up), the free tier of the Gemini API, or a local model on your own computer with Ollama or LM Studio.
Does ShroomPen charge anything on top?
No. ShroomPen is free, has no subscription, and adds no fee or markup. Requests go straight from your browser to your provider, which bills you directly.