Comparisons
AI model comparison for business: prices, context and EU data processing (October 2026)
As of 9 October 2026, the most expensive flagship models cost USD 10 per million input tokens and USD 50 per million output tokens, while the cheapest models for high volumes cost USD 0.10 and USD 0.50. The choice, however, comes down to three things: quality on your own data, any requirement for processing in the EU, and the licence, not the price alone.2749
Comparison table as of 9 October 2026
Prices are taken from the providers’ official websites and apply to standard processing. We update this comparison at least once a quarter, because price lists change often.
| Model | Provider | Price in USD per 1M tokens (input / output) | Context (tokens) | Processing in the EU | Licence / access |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic | 10 / 50 | 1M | provider API: no EU region; EU via regional cloud endpoints (+10%) | commercial API25 |
| Claude Opus 5.5 | Anthropic | 4 / 20 | 1M | provider API: no EU region; EU via regional cloud endpoints (+10%) | commercial API35 |
| Claude Sonnet 5.5 | Anthropic | 2 / 10 | 1M | provider API: no EU region; EU via regional cloud endpoints (+10%) | commercial API15 |
| Claude Haiku 5.5 | Anthropic | 0.10 / 0.50 (prompts up to 100K tokens) | 1M | provider API: no EU region; EU via regional cloud endpoints (+10%) | commercial API45 |
| GPT-6 Astra | OpenAI | 10 / 50 | 1.05M | Europe region (EEA + Switzerland), +10%, requires an additional agreement | commercial API710 |
| GPT-6.1 Sol | OpenAI | 2 / 10 | 1.05M | Europe region (EEA + Switzerland), +10%, requires an additional agreement | commercial API810 |
| GPT-6 Luna | OpenAI | 0.10 / 0.50 | 1.05M | Europe region (EEA + Switzerland), +10%, requires an additional agreement | commercial API910 |
| Gemini 3.1 Pro (preview) | 2 / 12 (prompts up to 200K tokens) | approx. 1M | Google Cloud: EU endpoint (multi-region) | commercial API1113 | |
| Gemini 3.8 Flash | 0.75 / 3.75 until 31.12.2026, then 1.50 / 7.50 | approx. 1M | Google Cloud: EU endpoint (multi-region) | commercial API121314 | |
| Mistral Large 4 (preview) | Mistral AI | 1.36 / 4.18 (promotional 0.68 / 2.09) | 1M | data in the EU by default | API; weights announced for the end of October151618 |
| Mistral Medium 3.5 | Mistral AI | 1.50 / 7.50 | 256K | data in the EU by default | open weights (modified MIT licence)1718 |
| Mistral Small 4 | Mistral AI | 0.15 / 0.60 | — | data in the EU by default | open weights, Apache 2.01618 |
| DeepSeek-V4.1-Flash | DeepSeek | 0.30 / 1.20 (peak hours) | 1M | API: data in China; weights can be run in-house | open weights, MIT licence192021 |
| Qwen3.6-35B-A3B | Alibaba (Qwen) | open weights only | 262K (up to approx. 1M) | run on your own infrastructure | open weights, Apache 2.022 |
How to read the prices
The price per million tokens is a unit cost, not the cost of an implementation. Batch processing cuts the price by 50% at Anthropic5 and by half at Google13. With some providers, long prompts cost more: for GPT-6 Astra the input rate doubles above 272K tokens7, and for Gemini 3.1 Pro it rises to USD 4 above 200K tokens13.
Promotions end on a set date: Gemini 3.8 Flash costs USD 0.75 per million input tokens until 31 December 2026, and from 1 January 2027 the price doubles13. Mistral Large 4 is in preview at a promotional price1516.
Data in the EU: what each provider offers
-
Mistral: data stored in the EU by default18.
-
OpenAI: a Europe region (EEA and Switzerland) for storage and processing, with a 10% surcharge for newer models10.
-
Google: in Google Cloud, the EU endpoint covers storage and processing in the member states14.
-
Anthropic: the provider’s own API offers only the “us” and “global” locations6; an EU region is available through regional cloud endpoints, at a 10% surcharge5.
-
DeepSeek: the API processes data in China21; for company data from the EU, the practical option is to run the open weights in-house20.
How to choose a model in practice
We start with a test set: several dozen real cases from your company, each with the expected result. On that set we compare two or three models from different price tiers and calculate the cost per task, not per token. A mid-range model often delivers a good enough result at a fraction of the flagship’s cost, but that has to be measured, not assumed.
Frequently asked questions
Which AI model is best for a business?
There is no single winner. We choose a model using a test set built from the company's real cases, checking quality, running costs and data requirements. Flagship models usually come out ahead for complex agentic tasks, and economy models for classification and high volumes.
Can the data stay in the European Union?
Yes, but it depends on the provider: Mistral stores data in the EU by default, OpenAI offers a Europe region at a 10% surcharge, and Google an EU endpoint in Google Cloud. For Claude, an EU region is provided through cloud platforms, at a 10% surcharge.1810145
Can prices per token be compared directly?
Only as a rough guide. Models split text into tokens in different ways; according to Anthropic, the newer Claude tokenizer produces around 30% more tokens for the same text. We measure the real cost on your documents.5
Sources
- Anthropic: Models overview
- Anthropic: Claude Fable 5.1
- Anthropic: Claude Opus 5.5
- Anthropic: Claude Haiku 5.5
- Anthropic: Pricing
- Anthropic: Data residency
- OpenAI: GPT-6 Astra
- OpenAI: GPT-6.1 Sol
- OpenAI: GPT-6 Luna
- OpenAI: Data controls in the OpenAI platform
- Google: Gemini 3.1 Pro Preview
- Google: Gemini 3.8 Flash
- Google: Gemini Developer API pricing
- Google Cloud: Data residency — Gemini Enterprise Agent Platform
- Mistral AI: Introducing Mistral Large 4
- Mistral AI: API pricing
- Mistral AI: Mistral Medium 3.5 (26.04) model card
- Mistral AI: Where do you store my data?
- DeepSeek: Models & Pricing
- DeepSeek (Hugging Face): DeepSeek-V4.1-Flash model card
- DeepSeek: Privacy Policy
- Qwen / Alibaba (Hugging Face): Qwen3.6-35B-A3B model card