AI prices
AI bills come as a free cap, a monthly price per user, or a meter that charges per million tokens. We read every price below on the vendor's own page and wrote the date beside it. Prices change often, so the page also shows how to redo the sums yourself, in dollars and in kronor.
Price changes this month
Beginner
Free, a monthly seat, or a token meter
AI bills come in three shapes. A free plan costs 0 dollars and stops you when you hit a cap. On 20 October 2026, Claude's free plan includes chat, web search, files and code, with usage limits that reset on a rolling five-hour window. Google's Gemini API free tier gives free input and output tokens on some models, and Google says that free-tier content may be used to improve its products. A flat monthly plan is one price for a bigger cap. Claude Pro is 20 dollars a month, or 17 dollars a month if you pay 200 dollars up front for the year. Those figures exclude tax. Pay-per-use charges for tokens. A token is a small piece of text the model reads or writes. Input tokens are what you send. Output tokens are what the model sends back.
A chat inside a subscription cap adds 0 dollars. The same chat on an API has a token bill. As a worked example, take 100 emails of 500 input tokens and 200 output tokens each, then replace those numbers with a count from your own account. Together that is 50,000 input tokens and 20,000 output tokens, or 0.05 million and 0.02 million. At Claude Haiku 5.5 short-prompt rates of 0.10 and 0.50 dollars per million, the bill is 0.015 dollars before tax. At Sonnet 5.5 rates of 2 and 10 dollars per million, it is 0.30 dollars. Write the date beside every price.
Price a month of small chats
Do not invent prices. I will paste prices and the date I read them. A token is a piece of text the model reads or writes. Ask me for chats per month, average input tokens, and average output tokens. Convert each side to millions of tokens. Monthly dollars before tax = input millions times the input price, plus output millions times the output price. Then compare that total with a monthly subscription price I paste. Show the chat count where the API total crosses the subscription. Leave a blank line labeled kronor, for me to fill after I look up the day's dollar rate.
Intermediate
Price per user, price per token, kronor and VAT
A monthly seat buys a usage cap. An API buys tokens, the small text pieces a model reads and writes, priced per million. On 20 October 2026 Claude Team is 20 dollars per standard seat per month when billed annually, and 25 dollars when billed monthly. A premium seat is 100 or 125 dollars, with about five times the usage of a standard seat. OpenAI's Sweden business page lists 180 kronor for a standard seat and 900 kronor for a premium seat, and equates them to 20 or 25 dollars and 100 or 125 dollars. Claude draws website, desktop, phone and Claude Code from one pool, and publishes no fixed message count.
Claude Sonnet 5.5 is 2 dollars per million input tokens and 10 per million output tokens. A cache read of a repeated prefix is 0.10 dollars per million. Batch, for jobs that can wait, is 1 and 5. Grok 4.7 under 200,000 prompt tokens is 2 input, 0.50 cached input and 6 output. Gemini 3.8 Flash paid standard is 0.75 and 3.75 through 31 December 2026, then 1.50 and 7.50 from 1 January 2027, with batch and flex at half those rates.
Swedish standard VAT is 25 percent of the price excluding VAT. A gross price is the net price times 1.25, and the VAT inside a gross price is 20 percent of the gross price. Dollar API invoices often print no Swedish VAT. Multiply by the Riksbank dollar rate for that day and store the date with the rate.
Compare one seat with one API job
Do not invent prices. I will paste a monthly seat price and an API card with input, output, cache-read and batch prices per million tokens, plus the date. Ask me how many requests I send in a month, the input tokens per request, and the output tokens per request. Compute the standard API total and the batch total. Say how many requests fit inside the seat price. Then ask me for a Riksbank dollar rate and a date. Show kronor excluding VAT, and kronor including 25 percent VAT using gross = net times 1.25. Label both results with the rate date.
Expert
Every price on a model, image prices, and rented GPUs
For each model, note every price that applies. Log input, output, cache read, cache write and batch, in dollars per million tokens, with the date and any prompt-length step. On 20 October 2026 Claude Opus 5.5 is 4 dollars of input and 20 of output per million tokens. A 5-minute cache write is 5 dollars per million and a cache hit is 0.20. Batch processing saves 50 percent, so the same model is 2 and 10.
OpenAI lists gpt-6-luna, at or under 272,000 tokens of context, at 0.10 input, 0.50 output and 0.01 cached input. gpt-6-astra is 10, 50 and 1. Its batch column is half those rates. Gemini 4 Argon paid standard is 2 and 10 through 30 November 2026 and 4 and 20 from 1 December 2026, with batch at half. xAI prices grok-imagine-image at 0.02 dollars per image, while OpenAI prices gpt-image-2 image tokens at 8 dollars per million input and 30 per million output.
Vast.ai bills GPUs by the second. On-demand runs keep running. Interruptible runs are listed as at least 50 percent cheaper and can be reclaimed. Reserved terms of 1, 3 or 6 months are listed as up to 50 percent off. Copy the live rate for a named GPU and date it.
Fill one price-card row
Do not invent prices. I will paste one vendor pricing page. Extract a table with columns: model, date read, prompt-length rule, input per 1M tokens, output per 1M, cache read, cache write, batch input, batch output, and any separate image unit. If a cell is missing on the page, write missing. Add a column for a scheduled future price if the page prints an effective date. Then list three lines I should log every day for 30 days before calling a dip a lasting price: the API row, one GPU hourly card with the instance type, and the URL.
Prices and the date we read them
| Item | Figure | What the meter measures | Read |
|---|---|---|---|
| Claude Free | $0 | Cap that resets on a rolling five-hour window | 20 Oct 2026 |
| Claude Pro | $20 a month, or $17 a month when $200 is paid up front | Larger cap. Tax extra | 20 Oct 2026 |
| Claude Team, standard seat | $20 a month billed annually, $25 billed monthly | Per seat | 20 Oct 2026 |
| Claude Team, premium seat | $100 annually, $125 monthly | About five times standard-seat usage | 20 Oct 2026 |
| OpenAI Business, standard seat | 180 kronor a month on the Sweden page, equated there to $20 annually or $25 monthly | Per seat | 20 Oct 2026 |
| OpenAI Business, premium seat | 900 kronor a month, equated to $100 annually or $125 monthly | Per seat | 20 Oct 2026 |
| Claude Haiku 5.5, prompt up to 100,000 tokens | $0.10 input and $0.50 output per million tokens | Tokens | 20 Oct 2026 |
| Claude Sonnet 5.5 | $2 and $10 per million; cache hit $0.10; batch $1 and $5 | Tokens | 20 Oct 2026 |
| Claude Opus 5.5 | $4 and $20 per million; 5-minute cache write $5; cache hit $0.20; batch half | Tokens | 20 Oct 2026 |
| Claude Fable 5.1 | $10 input and $50 output per million tokens; cache write $12.50; cache hit $0.25 | Tokens | 29 Oct 2026 |
| Grok 4.7, prompt under 200,000 tokens | $2 input, $0.50 cached input, $6 output | Tokens | 20 Oct 2026 |
| Gemini 3.8 Flash, paid standard | $0.75 and $3.75 per million through 31 December 2026, then $1.50 and $7.50 | Tokens. Batch and flex are half | 20 Oct 2026 |
| Gemini 4 Argon, paid standard | $2 and $10 through 30 November 2026, then $4 and $20 | Tokens. Batch is half | 20 Oct 2026 |
| grok-imagine-image | $0.02 per image | Image | 20 Oct 2026 |
| gpt-image-2 | $8 input and $30 output per million image tokens | Image tokens | 20 Oct 2026 |
| gpt-6-luna, context at or under 272,000 | $0.10 input, $0.50 output, $0.01 cached input | Tokens. Batch is half | 20 Oct 2026 |
| gpt-6-astra, short context | $10 input, $50 output, $1 cached input | Tokens. Batch is half | 20 Oct 2026 |
| Inception Mercury 3 | $0.20 input and $0.70 output per million; $0.10 and $0.35 at launch; cached input 90% off | Tokens | 29 Oct 2026 |
A 1,000-email basket, VAT, and a GPU worksheet
Teaching basket, one request per email so each prompt stays on the short-prompt rate: 1,000 emails, 500 input tokens and 200 output tokens each. That is 0.5 million input tokens and 0.2 million output tokens. Before tax, on the 20 October 2026 cards:
- Haiku 5.5: 0.5 times 0.10 plus 0.2 times 0.50 = 0.15 dollars.
- Sonnet 5.5: 0.5 times 2 plus 0.2 times 10 = 3.00 dollars. Batch half is 1.50 dollars.
- Grok 4.7 under 200,000 prompt tokens: 0.5 times 2 plus 0.2 times 6 = 2.20 dollars.
- Gemini 3.8 Flash paid standard, rate in force through 31 December 2026: 0.5 times 0.75 plus 0.2 times 3.75 = 1.125 dollars. Batch half is 0.5625 dollars.
Replace 500 and 200 with a count from your own logs. Kronor excluding VAT = dollars times the Riksbank dollar rate for that day. Gross at the standard 25 percent rate = that net times 1.25. VAT already inside a gross price is 20 percent of the gross price. For a graphics card you buy, the monthly cost per gigabyte of VRAM is the shop price including VAT, divided by its gigabytes of VRAM, divided by the months you plan to keep it. This page has no shop prices. Vast.ai hourly rates change all the time, so copy the live card and date it.
Log columns: date, vendor URL, model id, input, output, cache read, cache write, batch, currency, and for a GPU the instance type plus dollars per hour. Gemini has already printed two step-ups, on 1 December 2026 for Argon and on 1 January 2027 for the 3.8 Flash rates above, so a single screenshot will not show the next price.
What a shared price board will collect
A price post in the forum should carry the vendor URL, the read date, and the five token meters, or the GPU instance type and the hourly rate. Until such posts exist, the worksheet above is the record. A one-day promo stays in the log as a dip. A note that a rate has fallen waits until 30 daily rows exist for that same model id.
Hrafninn Industries AB plans to raise money for a DGX Station in Sweden that would run open-weight models. Read about the Station
Sources
- Inception: Introducing Mercury 3List price $0.20 input and $0.70 output per million tokens, half price at launch, cached input 90 percent off. Read 29 October 2026.
- Claude pricingFree, Pro, Max and Team seat prices, plus API cards for Opus, Sonnet and Haiku, cache rates, and the 50 percent batch saving. Read 20 October 2026; Fable 5.1 read 29 October 2026. Tax extra.
- Gemini Developer API pricingFree and paid tiers, Gemini 3.8 Flash and Gemini 4 Argon standard, batch and flex rates, and the dated price steps in December 2026 and January 2027.
- xAI models and API pricesGrok text prices per million tokens, including cached input and the 200,000-token step, plus per-image Imagine prices.
- OpenAI API pricingStandard, batch and cache prices per million tokens for gpt-6-luna, gpt-6-astra and others, plus gpt-image-2 image-token rates.
- OpenAI business pricingBusiness seat prices as shown for Sweden on 20 October 2026: 180 and 900 kronor, with the dollar equivalents printed on the same page.
- Skatteverket: how to charge VATStandard rate 25 percent. Net times 1.25 is the gross price. VAT inside a gross price at that rate is 20 percent of the gross price.
- Sveriges Riksbank exchange ratesLook up the dollar rate for the day you convert. This guide does not lock in a rate.
- Vast.ai GPU pricingOn-demand, interruptible (listed as 50 percent or more cheaper) and reserved terms (listed as up to 50 percent off). Hourly rates are live, so copy the card and date it.