AI News

Alibaba Releases Qwen4-Max Foundation Model, Advancing Agent and Coding Capabilities

LeiphoneUpdated Nov 16 21:03

AI summary

Alibaba has released its new-generation foundation model Qwen4-Max, with 2.4T total parameters and only 71B activated, featuring native vision capabilities and support for a million-token context. The model ranks 6th globally in the CodeArena front-end coding blind test and scored 46.7 on Agents' Last Exam, an improvement of 18.8 points over the previous generation. Qwen4-Max is now available on the Qwen AI platform as an API service, with cache hits costing as low as 0.5 yuan per million Tokens, and the average cost per task reduced by 58% compared with the previous generation. Small and medium-sized models will be open-sourced successively.

Why read this

AI score 90

The original article provides the new model's architecture parameters, multiple benchmark comparisons, and real-world deployment data, which can be used to assess its capability positioning and cost changes.

Topics

Tags

Body · AI translation

On November 16, Alibaba released Qwen4-Max, a new-generation foundation model with 2.4T total parameters, native vision capabilities, and support for million-token contexts. Built on an innovative model architecture and training optimizations, Qwen4-Max achieves higher performance with lower token consumption and continues to advance in key capability areas such as agents and coding: on the authoritative global front-end coding blind test CodeArena, Qwen4 ranks sixth; in third-party evaluations of agentic capabilities and professional tasks such as Agents' Last Exam (ALE) and Vals Index, Qwen4 is in the leading tier. Qwen4 is deeply optimized for real-world tasks across all domains, can better invoke professional tools to handle long-horizon tasks, and is advancing the deep deployment of AI in professional fields such as automotive manufacturing, game development, financial investment research, and scientific research.

Qwen4-Max is now available on the Qianwen AI platform, providing API services at a cache-hit price as low as 0.5 yuan per million tokens, and the average cost of completing a single task is down 60% from the previous-generation model; users can also call the new model through Alibaba Token Plan or experience it directly in applications such as Qianwen Office, Qoder, and the Qianwen APP. It is reported that the small- and medium-sized models in the Qwen4 series will be open-sourced and released successively.

(Caption: In the CodeArena global front-end coding leaderboard, Qwen4-Max ranks 6th)

Qwen4-Max adopts a new MoE architecture based on Qwen Sparse Attention (QSA), with only 71B of its 2.4T total parameters activated, 25% fewer activated parameters per token than the previous-generation model, and higher thinking efficiency: in the Terminal-Bench 4.0 professional coding evaluation, at the same Xhigh thinking level, Qwen4-Max's average token usage per attempt dropped from 289K to 151K compared with the previous-generation Qwen3.8-Max, a reduction of nearly 50%, while its evaluation score rose from 38.9 to 48.5, an improvement of nearly 25%. At the same time, thanks to the sharp drop in cache-hit pricing, evaluation results show that Qwen4-Max's average cost per task (Cost per Task) is 58% lower than that of Qwen3.8-Max.

The new architecture, combined with efficient training optimizations, has comprehensively improved Qwen4-Max's performance. In the CodeArena global blind-test authoritative leaderboard focused on coding, Qwen4-Max ranks 6th worldwide in front-end (Webdev), between GPT6-Astra and Claude Fable5.1, setting a new best result for Chinese models; on the "Agents' Last Exam" (ALE), it achieved the highest score of 46.7, a substantial improvement of 18.8 points over Qwen3.8-Max; in the Chartography evaluation, which assesses large models' ability to understand professional charts and reason, Qwen4-Max scored 54.2, up 12.2 points from the previous-generation model; in the JobBench professional work evaluation, the new model scored 68.5, surpassing GPT-6.1sol and trailing Claude Opus5.5; and on HLAB-Vals-1.1, an evaluation focused on the legal profession, Qwen4-Max scored 21.3, an improvement of 5.9 points over Qwen3.8-Max, placing it among the industry's leading models.

Whether for simple everyday needs or long-horizon complex tasks, Qwen4 can deliver reliably, making human-machine collaboration smoother and more efficient. Qwen4-Max implements a Live Delivery mechanism within the model, which can first generate a usable initial version and then continuously expand the content version by version according to user feedback until completion: for a design request for a physics classroom popular-science website, Qwen4-Max delivered 14 usable versions, ultimately expanding it into a polished website featuring 3D animations, interactive experiments, and detailed explanations of principles. At the same time, Qwen4-Max also possesses the ability to perform operations research modeling, solving, and verification around complex objectives and constraints, and through autonomous planning and continuous iteration, orchestrates agents to solve more complex tasks: based on Alibaba Cloud's AI infrastructure, Qwen4-Max autonomously and closed-loop optimized the operating system that supports its own applications, inference, and training, reducing model costs by 10% in data annotation scenarios and increasing the optimal solving speed in CPU offline rescheduling scenarios by more than 14 times.

Caption: Across 721 directions of real work domains covering all scenarios, Qwen4 improves overall professional capability by more than 10% compared with the previous-generation model

From professional collaboration (Cowork) to all-weather agents, Qwen4 is penetrating more real work tasks and production environments. The Qianwen team built the Work360 all-scenario modeling system covering 721 subdomains, and in systematic evaluations, Qwen4-Max scored 52.69, an improvement of more than 10% over the previous-generation model. In professional design fields such as game development, Qwen4 demonstrates strong 3D modeling capabilities and can better invoke professional tools such as Blender, generating a fun and cool 3D game with a single sentence. In the automotive manufacturing industry, Qwen4 has been applied throughout the entire process of production line planning, part design, and machining verification, slashing the manual time for machining programming of a single workpiece from 475 minutes to 24 minutes, a reduction of 95%; in financial investment research scenarios, Qwen4 proficiently invokes 90 MCP professional tools to produce Excel workpapers and PPT reports, supporting dual traceability to ensure reliable conclusions.

In the frontier field of AI-assisted scientific discovery (AI For Science), Qwen4 has initially produced real results in materials and drug discovery. In the field of materials discovery, Qwen4 ran autonomously for 48 consecutive hours, selected candidate molecules from 2.7 million candidates, and through four rounds of a "design—synthesize—test" closed loop, actually synthesized a new type of coolant material, providing a new approach to alleviating the data center cooling challenges brought by high-power chips; in the field of drug discovery, targeting a key target for Alzheimer's disease, Qwen4 completed 3,239 computational evaluations through 454 tool calls, designed and optimized 24 candidate molecules, and the final selected molecule's nominal binding affinity to the target was about 3.6 times higher than that of the starting molecule, making it a more effective new option.

(End)

Source

Leiphone