律动BlockBeats
律动BlockBeats|Nov 16, 2026 02:50
[Qwen4-Max Officially Released: Remaining Backbone Size, 25% Reduction in Activation Parameters, and 62% Drop in Task Cost] Beating AI News: Following last week's release of the preview version, Alibaba officially released its new flagship model, Qwen4-Max, announcing its complete technical architecture, benchmark scores, and API pricing. Qwen4-Max maintains the 2.4 trillion parameters of its predecessor's backbone, while introducing an additional 328 billion N-gram Embedding parameters. By redesigning the attention mechanism, residual connections, and phrase embedding layers, the backbone parameters activated per token generation have been reduced from approximately 95 billion to 71 billion, a reduction of approximately 25%. The model supports native visual understanding and 1 million token contexts. Official benchmarks show that Qwen4-Max scored 52.69 on the professional work benchmark Work360, compared to 47.74 for its predecessor. This Alibaba-built benchmark covers professional tasks in finance, law, manufacturing, and other fields. In Terminal-Bench 4.0 programming tests, the new model's token consumption decreased by approximately 48% to 49% compared to the previous generation. The average cost per task on Work360 dropped from $4.43 to $1.67, a decrease of 62%. Alibaba also showcased a research case study. Qwen4, in conjunction with an automated experimental platform, ran autonomously for 48 consecutive hours, screening novel coolant materials suitable for data centers from 2.7 million candidate molecules. After four rounds of molecular design, experimental synthesis, and testing, a physical sample of the candidate molecule was finally produced and validated experimentally. However, official data proving its actual cooling effect is superior to existing materials has not yet been released. In the automotive manufacturing case study, Qwen4-Max participated in part modeling, machining path planning, and simulation verification. The manual operation time for a workpiece machining programming task was reduced from approximately 475 minutes to 24 minutes, a reduction of approximately 95%. The model can also use Blender, Unreal Engine, and Unity to create 3D scenes and play games, and can be repeatedly modified based on operational feedback. Qwen4-Max has released its official API, supporting OpenAI and Anthropic compatible interfaces. The cost per million tokens is 10 yuan for input and 30 yuan for output, approximately 17% cheaper than the previous generation; the cost of a cache hit input has decreased from 1.5 yuan to 0.5 yuan. Alibaba has confirmed that smaller models in the Qwen4 series will be open-sourced later. [Original Link]
+3
Mentioned
Share To

Timeline

HotFlash

APP

X

Telegram

Facebook

Reddit

CopyLink

Hot Reads