Which is lower cost: 'cheap Chinese AI + human' or 'high-performance AI from OpenAI or Anthropic'?
While API usage fees for cutting-edge AI services such as
Outsourcing plus LocalAI will soon become more economical vs Frontier labs | SignalBloom AI posts
https://www.signalbloom.ai/posts/outsourcing-plus-localai-will-soon-become-more-economical-vs-frontier-labs/
As of the time of writing, the industry's top-performing AIs are American-made AIs developed by OpenAI, Anthropic, and Google. However, Chinese companies are also rapidly developing AI, and DeepSeek's DeepSeek-V4-Pro, released in May 2026, has been evaluated by a US government agency as having ' performance equivalent to GPT-5 and being 8 months behind American-made AIs .' Another feature of DeepSeek-V4-Pro is its lower API fees compared to American-made AIs, and on May 23, 2026, the company announced a permanent 75% discount, drawing even more attention to its high cost-performance ratio.
A 75% permanent discount on the high-performance Chinese AI 'DeepSeek-V4-Pro' is being offered, and attention is also being drawn to 'Reasonix,' a coding agent specifically designed for DeepSeek - GIGAZINE
The table below summarizes the API usage fees per million tokens for OpenAI, Anthropic, and DeepSeek. The input token prices are adjusted to reflect actual usage, including both cached and uncached inputs.
| input | output | |
| Anthropic | $1.57 (approximately 250 yen) | $25.00 (approximately 3985 yen) |
| OpenAI | $1.30 (approximately 207 yen) | $30.22 (approximately 4817 yen) |
| DeepSeek | $0.055 (approximately 8 yen) | $0.870 (approximately 139 yen) |
Based on the estimate that 'when using AI as a coding agent, the output is 50,000 tokens for every 1 million tokens input,' SignalBloom AI estimates the cost per 1 million tokens when using each company's API as a coding agent as follows:
Anthropic: $1.57 + $1.25 = $2.82 (approximately 450 yen)
OpenAI: $1.30 + $1.50 = $2.80 (approximately 446 yen)
DeepSeek: $0.05 + $0.0435 = $0.094 (approximately 15 yen)
Based on the premise that 'the number of tokens that AI users input into the AI increases over time,' SignalBloom AI has created a tool that allows users to compare the costs of a 'cutting-edge model' versus 'DeepSeek + human' by setting 'human engineer's salary,' 'salary increase rate (annual),' 'monthly number of tokens processed at the start of the estimate,' 'monthly token increase rate,' 'API fee for the cutting-edge model per 1 million tokens,' 'monthly API fee increase rate for the cutting-edge model,' and 'DeepSeek API fee per 1 million tokens.'
The following shows the cost progression assuming a human engineer's salary of $1,000 (approximately 160,000 yen), an annual salary increase rate of 8%, a processing capacity of 100 million tokens, a token growth rate of 8% per month, a cutting-edge model API fee of $2.89 (OpenAI's estimated price for article creation), a monthly price increase rate of 5% for the cutting-edge model, and a DeepSeek API fee of $0.094. The red line represents the cost of the 'cutting-edge model,' and the green line represents the cost of 'DeepSeek + human.' At the start of operation, 'DeepSeek + human' is more expensive, but after 11 months, the 'cutting-edge model' becomes more expensive, and from there the cost of the 'cutting-edge model' continues to rise.
This is what happens when you change the salary of a human engineer to $2000 (approximately 320,000 yen). After 17 months, the 'state-of-the-art model' becomes more expensive.
The tool in the SignalBloom AI article allows you to freely adjust the values of each parameter, so feel free to try it out if you're interested.
Related Posts:
in AI, Posted by log1o_hf