In a significant update, DeepSeek, Chinese AI startup, has officially unveiled its advanced flagship model, V4-Flash, which researchers say is currently the cheapest well-known AI model to run.
The launch comes as the company aims to outdo other leading players of the industry while reportedly preparing for a potential domestic initial public offering (IPO).
As per research firm Artificial Analysis, DeepSeek's V4-Flash costs an average of only 3 cents benchmark test, making it over 100 times cheaper to operate as compared to Anthropic’s Claude Fable 5, which averages $3.15 per test.
The recently introduced model undercuts OpenAI’s GPT-5.6 Solu at $1.86 per test and Moonshot AI's Kimi K3 at 86 cents.
The pricing is dependent on token usage, with DeepSeek charging $0.14 per million input tokens and $0.28 per million output tokens.
Scientists mentioned that benchmark costs offer a more realistic comparison as compared to pricing alone because they account for the processing amount needed to complete tasks.
DeepSeek new V4-Flash tool: What you must know?
Despite its affordability, V4-Flash trails some rivals in performance. On Artificial Analysis' Intelligence Index that efficiently evaluates models across coding, reasoning and workplace-related tasks, the model scored 50 out of 100.
It matched Google's Gemini 3.6 Flash; however, it ranked behind Moonshot's Kimi K3, which scored 57, while Anthropic's latest models and OpenAI's GPT-5.6 achieved even higher scores.
For those unaware, DeepSeek captured tremendous attraction in early 2025 with its low-cost R1 model, but has since experienced severe competition from Chinese rivals including Moonshot AI, MiniMax, Z.AI, ByteDance and Alibaba.
Moreover, the company is developing the cutting-edge V4-Pro model, but its release date has yet to be officially confirmed.
Meanwhile, Alibaba has launched its newest Qwen3.8-Max AI model, igniting competition in China’s evolving AI industry.
