DeepSeek’s V4-Flash has set a new low-cost benchmark among widely known AI models, according to research firm Artificial Analysis. The model costs an average of 3 cents per benchmark test, compared with $3.15 for Anthropic’s Claude Fable 5.

The pricing gap highlights how Chinese AI companies are competing with Western developers on operating costs while reducing differences in measured model performance. DeepSeek’s approach centers on lower-priced access, while Alibaba has introduced its largest AI model so far, Qwen3.8-Max.

DeepSeek V4-Flash Pricing and Performance

DeepSeek prices V4-Flash at $0.14 per million input tokens and $0.28 per million output tokens. Artificial Analysis estimated that its cost per task is about 60 times lower than OpenAI’s GPT-5.6 Sol, which was measured at $1.86 per test, and more than 100 times lower than Claude Fable 5.

The benchmark comparison goes beyond published token prices. It considers the volume of data a model must process and generate to complete a task, providing a task-based view of operating cost.

V4-Flash Benchmark Results

On the Artificial Analysis Intelligence Index, V4-Flash received a score of 50 out of 100. That result matched Google’s Gemini 3.6 Flash and placed the model one point behind Meta’s Muse Spark 1.1.

V4-Flash uses a 284-billion-parameter mixture-of-experts design. Only 13 billion parameters are active for each token, a structure associated with its lower operating costs.

Alibaba Launches Qwen3.8-Max

Alibaba introduced Qwen3.8-Max on the same day the Artificial Analysis findings were published. The company described it as its largest AI model to date.

Qwen3.8-Max has 2.4 trillion parameters and supports a one-million-token context window. On the Arena.AI text leaderboard, it reached the highest position among Chinese models. It also ranked second globally on the organization’s vision leaderboard, behind Anthropic’s Fable 5 variants.

Alibaba’s Hong Kong-listed shares rose 7% following the news, according to Reuters.

Competition Among Chinese AI Companies Intensifies

DeepSeek is operating in a more crowded domestic AI market. Its competitors include ByteDance, Moonshot AI, MiniMax, and Z.AI.

Lian Jye Su, chief analyst at Omdia, said Chinese AI companies have identified demand for models that are affordable, transparent, accessible, and sufficiently capable, rather than models positioned solely as the industry’s top performers.

DeepSeek is also developing V4-Pro, a more capable successor to V4-Flash. No release date has been announced. Reuters has reported that the company is preparing for a potential initial public offering.

Why AI Model Cost Per Task Matters

Token pricing gives one view of AI model cost, but it does not show the full amount of work required to complete a benchmark task. A model may need to process and generate different volumes of data, which can change the practical cost of use.

Artificial Analysis’s cost-per-task measurement accounts for that difference. Under this comparison, V4-Flash’s 3-cent average cost stands far below the listed benchmark costs for Claude Fable 5 and GPT-5.6 Sol.

The result places DeepSeek’s low-cost model at the center of a broader AI market shift, where price, accessibility, performance, and model scale are all becoming key points of competition.