Cost against price
What does it actually cost to run?
Almost every AI tool rents its intelligence. The rent has a published price, which makes it the one part of a software bill you can work out exactly. Everything left over is the layer the company built.
431
tools in the index publish a monthly price
$20
is the median of those, per seat, per month
$0.13 to $48
the light workload, cheapest model to dearest
Work it out
Pick a shape of work and a model, then put in what you pay per seat. The arithmetic is the same one the vendor did, run in the other direction.
One person using an assistant through a working day. 20 calls a day, 4,000 tokens in and 800 out on each.
Light is the default because the unit here is one seat, and one person does not make a hundred calls a day. Reference and Heavy describe a whole feature at volume, so against a single seat price they will read as a loss on every row. That is the comparison being wrong, not the vendor.
One seat, one month, Gemini 3.1 Pro
- You pay
- $20
- The model costs
- $11
- Left for everything elseinterface, workflow, data, evaluation, support, profit
- $9.4447% of the price
This is a model, not an audit. We do not know which model any product runs on, how much work a seat really does, or what the vendor has negotiated. It excludes caching, batching and volume discounts, all of which push a real bill down. Treat the output as the size of the question, not as the vendor’s accounts.
By category
The median published price in each category, against the light workload on Gemini 3.1 Pro. Only categories where at least ten tools publish a monthly figure, because below that a median is one pricing page wearing a hat.
| Category | Priced | Median | Model cost | Left over | |
|---|---|---|---|---|---|
| Writing | 73 | $23 | $11 | $12 (54%) | Detail → |
| Video | 46 | $20 | $11 | $8.94 (46%) | Detail → |
| Coding | 44 | $20 | $11 | $8.94 (46%) | Detail → |
| Chatbot | 39 | $19 | $11 | $8.44 (44%) | Detail → |
| Productivity | 39 | $16 | $11 | $5.43 (34%) | Detail → |
| Image | 38 | $12 | $11 | $1.44 (12%) | Detail → |
| ML & infrastructure | 31 | $25 | $11 | $14 (58%) | Detail → |
| Audio | 30 | $25 | $11 | $14 (57%) | Detail → |
| Agents | 20 | $45 | $11 | $34 (76%) | Detail → |
| Search | 13 | $15 | $11 | $4.44 (30%) | Detail → |
| Marketing | 13 | $35 | $11 | $24 (70%) | Detail → |
| Data | 11 | $40 | $11 | $29 (74%) | Detail → |
Same model on every row, so the column compares categories rather than model choices.
The same job, priced 17 ways
Every priced model in the index, running the light workload. The choice of model moves the bill more than almost any other decision, and it is usually made once and never revisited.
| Model | Developer | In / out per M | Light a month |
|---|---|---|---|
| Qwen 3.7 Flash | Alibaba | $0.03 / $0.13 | $0.13 |
| DeepSeek V4 Flash | DeepSeek | $0.14 / $0.28 | $0.47 |
| DeepSeek V4 Pro | DeepSeek | $0.435 / $0.87 | $1.46 |
| GPT-5.6 Luna | OpenAI | $0.2 / $1.2 | $1.06 |
| Grok 4.20 | xAI | $1.25 / $2.5 | $4.20 |
| Claude Haiku 4.5 | Anthropic | $1 / $5 | $4.80 |
| Gemini 3.6 Flash | $1.5 / $7.5 | $7.20 | |
| GPT-5.6 Terra | OpenAI | $2 / $12 | $11 |
| Gemini 3.1 Pro | $2 / $12 | $11 | |
| Claude Sonnet 5 | Anthropic | $3 / $15 | $14 |
| Claude Sonnet 4.6 | Anthropic | $3 / $15 | $14 |
| GPT-5.4 | OpenAI | $2.5 / $15 | $13 |
| Claude Opus 5 | Anthropic | $5 / $25 | $24 |
| Claude Opus 4.8 | Anthropic | $5 / $25 | $24 |
| GPT-5.6 Sol | OpenAI | $5 / $30 | $26 |
| GPT-5.5 | OpenAI | $5 / $30 | $26 |
| Claude Fable 5 | Anthropic | $10 / $50 | $48 |