By Eduardo Baptista
BEIJING, Aug 3 (Reuters) – China’s Alibaba on Monday unveiled its largest and most capable AI model to date, sending its shares surging, while a research firm said DeepSeek’s latest product offers cut-throat pricing that is more than 100 times cheaper than Anthropic’s Claude Fable 5.
The two developments highlight the rapid pace of advancement in artificial intelligence by Chinese tech firms, which are locked in a fierce and fast-moving battle to build more powerful systems without making them prohibitively expensive to run.
Both models — Alibaba’s Qwen3.8-Max and DeepSeek’s V4-Flash — underline Chinese commitment to open-weight models as the firms seek to gain traction among developers globally.
“Chinese AI companies have found an important market. Many business workflows do not need the industry’s very best model,” said Lian Jye Su, chief analyst at research firm Omdia.
“They need models that are good enough, affordable, transparent and accessible, and open-weight models help meet that demand.”
With an open-weight model, the underlying learned settings that allow developers to run or adapt the system are available for download. By contrast, OpenAI, Anthropic and Google have closed-source models.
TRILLIONS OF PARAMETERS
Alibaba’s new Qwen3.8-Max immediately shot up leaderboards assessing the capabilities of AI models after being unveiled on Monday, helping its shares jump 7% in Hong Kong trade.
The model has 2.4 trillion parameters, the numerical settings a model learns from data and uses to recognise patterns, generate answers, and carry out tasks. That puts it not too far behind domestic rival Moonshot AI’s Kimi K3, which was launched last month and has 2.8 trillion parameters.
A higher parameter figure does not automatically make a model better, but it has become a closely watched measure of the scale of the computing and data behind advanced AI systems.
Qwen3.8-Max was unveiled on crowdsourced, model-comparison platform Arena.AI. It soon became the highest-ranking Chinese model in terms of text models, though it still lags Claude Fable 5 and three Opus variants which are all from Anthropic.
On Arena.AI’s leaderboard for AI models that analyse images and other visual material, Qwen3.8-Max ranked second globally, only behind a Claude Fable 5 variant.
Both Qwen3.8-Max and Kimi K3 can handle text, images and video, and process up to 1 million tokens at a time.
Tokens are chunks of data, often parts of words or short words, and a big figure means the model can take in large amounts of material in one go, such as long legal files, a large software codebase or hundreds of pages of documents.
The tech giant said the model, due to be released next week, completed a software-engineering project in 16 days. It uses a “mixture-of-experts” design, which divides work among specialised parts of the system instead of switching on the entire model for every request. Only 95 billion parameters are used at a time, reducing costs and response delays.
DEEPSEEK IS ULTRA CHEAP
DeepSeek’s V4-Flash model, released on Friday, is by far the least expensive to run on benchmark tests among well-known models globally, according to research firm Artificial Analysis.
The startup, which sources have said is preparing for a potential IPO, saw its R1 and V3 models become a global sensation in early 2025, triggering a selloff in global tech stocks and raising questions about the large amounts U.S. companies were spending on AI.
V4-Flash charges $0.14 per million input tokens and $0.28 per million output tokens, according to San Francisco-based Artificial Analysis.
Artificial Analysis estimated V4-Flash’s average cost at 3 cents per test, compared with 86 cents for Kimi K3, $1.86 for OpenAI’s GPT-5.6 Sol and $3.15 for Claude Fable 5.
The comparison provides a more realistic measure of value than pricing alone because it accounts for the amount of data a model must process and generate to complete a task. A model with low headline price can still prove expensive if it requires significantly more steps to produce an answer.
(Reporting by Eduardo Baptista; Editing by Edwina Gibbs)
Comments