---
title: "Chinese Tech Breakthrough: Alibaba Unveils Most Powerful AI Model, While DeepSeek Strikes with Prices"
description: "Chinese tech giants have made a powerful leap in the AI sector. Alibaba unveiled the Qwen3.8-Max model with 2.4 trillion parameters, leading to a 7% stock increase. Meanwhile, DeepSeek released the V4-Flash model, which is 100 times cheaper than analogues from Anthropic and OpenAI. 🇨🇳🤖📉"
date: 2026-08-04T00:16:58.000Z
lang: en
url: https://xab.info/en/posts/chinese-ai-models-alibaba-and-deepseek-new-market-leaders-en
tags: [alibaba, deepseek, artificial-intelligence, qwen, china-tech]
publisher: "XAB.info"
---

# Chinese Tech Breakthrough: Alibaba Unveils Most Powerful AI Model, While DeepSeek Strikes with Prices

![Letters QWEN on wooden Scrabble tiles, symbolizing Alibaba's AI model Qwen in the context of China's tech breakthrough](https://xab.info/media/2026/08/04/kitayskie-ii-modeli-alibaba-i-deepseek-novye-lideri-rynka/kitayskie-ii-modeli-alibaba-i-deepseek-novye-lideri-rynka-1.webp)

The Chinese technology sector has made a powerful leap on the global artificial intelligence stage. On Monday, Alibaba unveiled its largest and most powerful model to date, which immediately impacted financial indicators: the corporation's shares jumped 7% in Hong Kong trading. Meanwhile, competitor DeepSeek launched a model offering services more than 100 times cheaper than the flagship product of the American company Anthropic.

### New Giant from Alibaba: Qwen3.8-Max

The new Qwen3.8-Max model immediately secured high positions in international rankings. It supports 2.4 trillion parameters — numerical settings the system uses to recognize patterns and generate responses. In this regard, it lags only slightly behind the Kimi K3 model from Moonshot AI, released a month earlier (2.8 trillion parameters). Experts note that the number of parameters alone does not guarantee superiority, but serves as an important indicator of the scale of computing resources and data embedded in the system.

On the crowd-sourced platform Arena.AI, where neural network capabilities are compared, Qwen3.8-Max became one of the leaders among Chinese text models. In the category of image and visual material analysis, it took second place globally, trailing only the Claude Fable 5 model from Anthropic.

### Technological Efficiency and Speed

The key advantage of Qwen3.8-Max is its ability to work with massive volumes of data. The model can process up to 1 million tokens (data fragments) at a time. This allows it to analyze long legal documents, extensive codebases, or hundreds of pages of text simultaneously.

To achieve high performance, Alibaba utilized a "mixture of experts" architecture. It distributes tasks between specialized parts of the system, utilizing only 95 billion parameters simultaneously. This approach reduces costs and response latency. As proof of efficiency, the company cited an example: Qwen3.8-Max completed a software development project in just 16 days.

### The Battle for Accessibility: DeepSeek vs. Giants

While Alibaba bets on power, DeepSeek is attacking the market with price. Their V4-Flash model, released on Friday, is the cheapest among known models worldwide, according to research firm Artificial Analysis. The cost of using V4-Flash is $0.14 per million input tokens and $0.28 per million output tokens.

For comparison, the average test cost for V4-Flash is estimated at $0.03. Meanwhile, competitors offer services at significantly higher rates:

    - Kimi K3 — $0.86;

    - GPT-5.6 Sol from OpenAI — $1.86;

    - Claude Fable 5 — $3.15.

### Open Model Strategy

Both Chinese novelties demonstrate a commitment to an open model policy. Unlike OpenAI, Anthropic, and Google, which use closed-source code, Chinese companies provide developers with access to basic training parameters. This allows specialists worldwide to launch or adapt systems to their needs.

Lian Jye Su, Chief Analyst at research firm Omdia, notes that Chinese developers have found their niche: "For many business processes, the best-in-class model is not required. They need solutions that are good enough, affordable, transparent, and easy to use".