A DeepSeek AI model is outperforming its competitors in terms of cheapness, as new analysis shows that the model runs much cheaper than well-known rivals.

Benchmark AI performance analysis done by Artificial Analysis found that DeepSeeks newly released V4-Flash model cost about 3 cents per test, compared to OpenAI’s GPT-5.6 Sol cost of $1.86 and Claude Fable 5’s hefty $3.15. Chinese firm Moonshot AI firm’s model Kim K3 cost 86 cents per test.

V4 Flash was released on Friday as DeepSeek tried to garner the same excitement that the release of its R1 model – which shocked audiences with its apparent low-cost.

DeepSeek’s V4-Flash model charges 14 cents per million input tokens and 28 cents per million output tokens, the unit used to measure AI use.

Putting these figures to test, V4 Flash stayed consistently cheaper than its competitors when completing the same tasks by a significant margin – its average cost is about 100 times cheaper than Anthropic’s Claude model.

While DeepSeek was the first major Chinese developer to stir headlines with its lower cost performance, other Chinese competitors soon arose, such as Moonshot, as well as top tech firms like Alibaba and ByteDance launching their own models.

Recommended reading

• White House Warns of Chinese AI “Theft” as DeepSeek Releases New Model
• DeepSeek to Release New AI Model
• UK CISOs Call for DeepSeek Regulation to Avoid Cyber Crisis

The benchmark analysis from Artificial Analysis did not see DeepSeek completely come out on top, however. Its V4-Flash model scored 50 out of 100 points on the firm’s Intelligence Index, which benchmarks coding, assignment performance and reasoning ability, as well as cost.

This was in line with Google’s Gemini 3.6 Flash, and only one point away from Meta’s Muse Spark 1.1. Claude Opus and Fable 5 from Anthropic came out on top at 61 and 60 respectively, followed by OpenAI’s GPT-5.6 Sol (59).