A new version of Chinese startup DeepSeek’s flagship AI model has become the cheapest well-known AI model to run in benchmark tests, according to research firm Artificial Analysis. The firm said the model is more than 100 times cheaper to operate than Anthropic’s Claude Fable 5.
DeepSeek, which sources say is preparing for a possible initial public offering (IPO), officially launched its V4-Flash model on Friday. The company is once again focusing on what made it popular by offering powerful AI models at very low prices.
DeepSeek first gained worldwide attention in early 2025 with its R1 model. The launch triggered a sell-off in global technology stocks and raised questions about the large amounts of money U.S. companies were spending on artificial intelligence.
According to Artificial Analysis, DeepSeek’s V4-Flash charges $0.14 per million input tokens and $0.28 per million output tokens. Tokens are small units of data used to measure how much an AI model processes and generates.
The research firm estimated that V4-Flash costs about 3 cents per benchmark test. In comparison, Moonshot AI’s Kimi K3 costs about 86 cents per test, OpenAI’s GPT-5.6 Sol costs $1.86, and Anthropic’s Claude Fable 5 costs $3.15.
The comparison goes beyond basic pricing because it measures the total cost of completing a task. Even if an AI model appears cheaper at first, it can become more expensive if it needs to process or generate much more data before producing an answer.
DeepSeek was once the biggest name in China’s AI industry, but it now faces strong competition from local startups such as Moonshot, MiniMax and Z.AI, as well as major technology companies including ByteDance and Alibaba. These companies are also competing with U.S. AI firms by offering businesses more affordable AI solutions.
Artificial Analysis gave DeepSeek’s V4-Flash a score of 50 out of 100 on its Intelligence Index, which measures performance across nine benchmarks covering coding, reasoning and workplace-related tasks.
The model received the same score as Google’s Gemini 3.6 Flash and was one point behind Meta’s Muse Spark 1.1 and Z.AI’s GLM-5.2 model, developed by Zhipu.
However, Moonshot’s Kimi K3 scored 57, while Anthropic’s Claude Opus 5, Claude Fable 5 and OpenAI’s GPT-5.6 scored at least nine points higher than DeepSeek’s model.
DeepSeek is also working on a more advanced version called V4-Pro, although the company has not announced a release date.
Meanwhile, on Monday, Alibaba introduced its biggest and most advanced AI model to date, called Qwen3.8-Max. The company said the model is close in size to a competing AI model launched by Chinese startup Moonshot AI last month.
