Qwen4-Max Cuts Average Cost per Task by 60%

On November 16, Alibaba unveiled Qwen4-Max, its next-generation foundation model. The model has 2.4 trillion parameters, native vision capabilities and support for context windows of up to 1 million tokens. Through an innovative model architecture and training optimizations, Qwen4-Max delivers stronger performance while consuming fewer tokens.

Qwen4-Max is now available via API on the Qwen AI platform. Cached input costs as little as 0.5 yuan per million tokens, while the average cost of completing a single task is 60% lower than with the previous-generation model. Users can also access the new model through Alibaba's Token Plan or try it directly in Qwen Office, Qoder and the Qwen app. Alibaba will progressively release the small and midsize models in the Qwen4 series as open source.

Hunyuan Hy4 Flash: 12B Active Parameters, Input Priced at 0.5 Yuan

On November 16, Tencent released Hy4 Flash, the latest language model in its Hunyuan family. The model has 224 billion total parameters, including 64 billion Engram parameters, with 12 billion active parameters. It supports multimodal inputs and a context window of up to 1 million tokens.

Hy4 Flash costs 0.5 yuan per million input tokens and 2 yuan per million output tokens, falling to just 0.02 yuan for cached input. That is significantly cheaper than Hy4.

DeepSeek Imposes Usage Caps Amid Computing Power Shortage

On November 16, the STAR Market Daily reported that DeepSeek had imposed limits on long conversations in its app and web interface. Several users said DeepSeek displayed a notice stating that, due to “insufficient computing power, long conversations inactive for more than seven days are no longer supported.” The company also reduced the maximum length of a single text input to roughly 40,000 Chinese characters.

DeepSeek is also limiting daily conversation quotas. Once users reach the daily cap, they see a message stating: “Due to heavy server demand, your usage limit has been reached.”

Doubao Workspace Launches Across 21 Professional Scenarios

Doubao for Work has launched a workspace feature, initially covering 21 professional scenarios across finance, healthcare, education, creative design and other fields. The feature is now available in both the Doubao for Work desktop app and the standard Doubao desktop app.

According to Doubao for Work, Workspace is a dedicated AI environment designed for professional tasks. The skills and tools required for 21 scenarios spanning finance, healthcare, education and creative design come preconfigured, eliminating the need for users to select them individually.

AMD Acquires World Labs, Names Fei-Fei Li Executive Vice President

On November 16, AMD announced that it had completed its acquisition of AI research company World Labs, strengthening its R&D capabilities in AI models. World Labs focuses on AI models that can create and simulate interactive 3D environments, as well as technologies for robot learning and simulation.

World Labs founder Fei-Fei Li has joined AMD as executive vice president and chief scientist, reporting directly to AMD Chair and CEO Lisa Su.

TensorWave and OpenAI Sign Six-Year, 457 MW Deal

On November 16, US AI cloud provider TensorWave announced a six-year agreement with OpenAI to provide 457 MW of total IT capacity across multiple US data centers. The deployment will use AMD Instinct MI455X GPUs to supply dedicated AI computing power to OpenAI.

The facilities are purpose-built for AMD's Helios rack-scale systems. Computing power deliveries are expected to begin in the first half of 2027, with full deployment scheduled for early 2028.