The latest models undercut Western rivals on cost, scale, and accessibility

Two Chinese AI firms are intensifying the global AI price war by pairing faster model releases with sharply lower operating costs than their Western rivals.

DeepSeek’s V4-Flash, which entered public beta on 31 July, reportedly averages just three cents per benchmark test, versus US$3.15 for Anthropic’s Claude Fable 5, according to Artificial Analysis. That makes V4-Flash the cheapest widely known model to run, at least in that firm’s estimate, and it highlights how Chinese AI developers are pushing value as aggressively as capability. 

At the token level the pricing gap is especially striking: DeepSeek says V4-Flash costs US$0.14 per million input tokens and US$0.28 per million output tokens, while Artificial Analysis put its cost-per-task at roughly 60 times lower than OpenAI’s GPT-5.6 Sol, and more than 100 times below Claude Fable 5. The benchmark comparison is important because it reflects the actual amount of data a model must read and generate to complete a job, which can be more meaningful than a simple sticker price. 

Other performance advantages
Performance has not been sacrificed entirely for price. V4-Flash scored 50 out of 100 on its Intelligence Index, equal to Google’s Gemini 3.6 Flash, and just one point behind Meta’s Muse Spark 1.1. The model is built as a 284-billion-parameter mixture-of-experts system, but only 13bn parameters are active for each token — a design that helps keep operating costs low while preserving useful capability. Also:

  • Alibaba has added more fuel to the competition on the same day the DeepSeek pricing data circulated. It has unveiled Qwen3.8-Max, its biggest AI model yet, with 2.4tn parameters and a one-million-token context window. The model has risen to the top Chinese position on the Arena.AI text leaderboard and reached second place globally on the vision leaderboard, behind only Anthropic’s Fable 5 variants, while Alibaba’s Hong Kong-listed shares climbed 7% after the announcement. 
  • The broader market is clearly moving towards a more crowded and price-sensitive phase. Reuters reported that DeepSeek is preparing for a possible initial public offering, but the firm now faces stronger domestic competition from ByteDance, Moonshot AI, MiniMax, and Z.AI.
  • DeepSeek is also working on V4-Pro as a more capable successor, although no launch date has been set.

One Omdia analyst has said Chinese AI firms are targeting customers who want models that are “good enough, affordable, transparent and accessible,” suggesting the fight is increasingly about practical utility rather than chasing the absolute frontier.