Three AI models, three price tiers: DeepSeek is the cheapest

If you build AI into a product, you typically pay per token. A token is a small piece of text, and prices are quoted per million tokens. Here is what three providers charge, according to the available pricing pages and reviews. All prices are in US dollars.

DeepSeek: price depends on time of day

DeepSeek-flash costs $0.15-0.30 per million input tokens and $0.60-1.20 for output. The more expensive deepseek-v4-pro costs $0.66-1.32 for input and $1.98-3.96 for output. If the input has already been used before and stored in an intermediate cache, it is far cheaper.

The price depends on the time of day. The high end of the range, for example $1.20 for flash output, applies during peak load. The low end, for example $0.60, applies outside peak load, when it costs half as much. Peak load is Monday-Friday 01-04 and 06-10 UTC, except Chinese public holidays.

DeepSeek-flash is currently served by DeepSeek-V4.1-Flash-1017 and can understand images. Deepseek-v4-pro cannot.

Mistral and Cohere

According to the review from eesel.ai, Mistral Large 4 has a list price of $1.36 for input and $4.18 for output. There is a 50 percent launch discount, bringing that to $0.68 and $2.09. Eesel.ai gives no end date, and the sources do not make clear whether the discount still applies. Mistral's own pricing page was not used, so the figures come from a third party.

The model has a context window of 1 million tokens and is in public preview. Open weights are expected by the end of October 2026.

According to eesel.ai, about 75 percent of the model's output is internal reasoning that the user does not see. As a result, Mistral Large 4 costs roughly $0.57 per benchmark task at the discounted launch price. That is about twice as much as DeepSeek V4.1 Flash for comparable performance, according to eesel.ai.

Cohere Command A costs $2.50 for input and $10.00 for output, according to the price aggregator Price Per Token. Cohere's own pricing page does not list Command A, only Command R+ and Aya Expanse.

What the difference means

Based on output prices, Command A is about 8-9 times more expensive than deepseek-flash at its top price and about 2.4 times more expensive than Mistral Large 4 at list price. Compared with flash's lowest price, the difference is about 16.7 times. This is a calculation based on the sources' figures.

Token prices don't tell the whole story about what a task costs, however, especially for models that do a lot of thinking along the way.