Business and economy·October 24, 2026, 14:14

Cheap AI on paper: Where the DeepSeek and Claude comparison falls short

AI-generated from the sources listed, with automated source checks.

A widely shared comparison pits DeepSeek's $0.15 against Claude's $4 per million tokens. DeepSeek's $0.15 matches its own pricing page, while Claude's $4 is only indirectly supported. The figures cover different models, times and types of usage.

A headline on the site tech-insider.org sets $0.15 against $4 per million tokens and makes it sound as if DeepSeek is significantly cheaper than Claude. Tokens are the small chunks of text an AI model works with when it reads and writes. The price depends on how many are used.

What the $0.15 covers

According to DeepSeek's own pricing page, $0.15 is the price for sending text in (input) to the deepseek-flash model. It applies only outside peak hours and without a cache hit, meaning when the model cannot reuse earlier text. During peak hours the price is $0.30.

Peak hours are 01-04 and 06-10 UTC Monday through Friday, except Chinese public holidays. So it is not a general price for DeepSeek. The larger model, deepseek-v4-pro, costs $0.66 for input and $1.98 for output outside peak hours. Output is what the model writes back. For deepseek-flash, the output price is $0.60.

And the $4 at Claude

Tech-insider compares it with a Claude model the article calls Opus 5.5, at $4 for input. Anthropic's pricing page indirectly supports the figure: A cache hit on Opus 5.5 is $0.20, which is 5 percent of the input price, corresponding to $4. The actual table price does not appear in the excerpt of the page that is shown, however.

The displayed portion of Anthropic's page instead mentions Claude Fable 5.1 and Mythos 5.1 at $10 for input and $50 for output.

According to tech-insider, Haiku 5.5 costs $0.10 for input and $0.50 for output for prompts of up to 100,000 tokens. That would make it cheaper than DeepSeek on short prompts. However, those prices appear only in tech-insider and are not found in the displayed excerpt of Anthropic's page.

Price per token is not price per task

Tech-insider itself argues that you should look at the price per completed task. According to the article, DeepSeek is more verbose and uses more tokens, while Claude solves tasks with fewer. That is the article's assessment and is not backed by figures.

The comparison in the headline thus mixes different models and times. Prices also change often, and the figures here apply as of October 24, 2026.