11-19 更新 实时
首页/ AI 大模型/ 正文

阿里 Qwen4-Max 低价对标 Opus 5.5、GPT-6 与 Kimi K3.1 Pro

AI 大模型 来源:The Decoder 2026-11-17 00:56 0 阅读 机器翻译

阿里巴巴发布 Qwen4-Max,其重构后的架构能以更高效率、更少算力处理长文本。

在 Artificial Analysis Intelligence Index 上,该模型排名第二,仅次于 Kimi K3.1 Pro,但每任务成本约为其前代的三分之一。

一项名为“Live Delivery”的功能让用户可以提前查看自主智能体产生的中间结果,并随时调整正在运行的任务。

Qwen4-Max 是阿里巴巴对 Opus 5.5、GPT-6 和 Kimi K3.1 Pro 的回应。这一新旗舰将更高效的架构与大幅降低的价格结合在一起,面向工业、科研和金融领域的专业工作。

据 Qwen 团队介绍,这款混合专家模型拥有 2.4 万亿参数,但每个 Token 仅激活 710 亿参数。它可原生处理图像,并提供 100 万 Token 的上下文窗口。

混合架构让长上下文的处理成本更低

Qwen4-Max 首次采用 Qwen4 系列重构后的架构。阿里巴巴不再仅依赖标准注意力层,而是将两种方法结合起来:三层将先前的上下文压缩为一种浓缩状态,第四层则只挑出真正重要的部分。这大幅削减了长输入的计算成本,因为模型不再需要对上下文的每一部分给予同等关注。

Qwen4-Max 先用三层压缩先前的文本,再由第四层只对重要部分进行细看。| 图片来源:Alibaba

该架构还在模型层之间增加了四条并行连接,并配有动态门控,用于稳定信息流动。一个拥有 3280 亿参数的 N-gram 嵌入层提供常见 Token 序列,作为现成的构建模块。据阿里巴巴称,2.4 万亿参数配置的训练运行稳定。整个训练过程中训练损失稳步下降,没有出现突然的尖峰。

据阿里巴巴称,整个训练过程中训练损失稳步下降,没有出现损失尖峰。 | 图片:阿里巴巴

阿里巴巴的内部评估以每字节比特数(bits per byte)衡量模型在知识、代码、问答和 STEM 方面的质量。在测试的五个检查点中,Qwen4-Max-Base 得分最佳。相比 Qwen3.8-Max-Base,它提升了 5% 至 13%,同时每个 Token 激活的参数减少了 25%。

Qwen4-Max-Base 在全部四个类别中都取得了最低、因而也是最佳的数值。 | 图片:阿里巴巴

更低的 Token 用量使每项任务的成本降低一半以上

效率提升体现在运营成本上。Work360 是阿里巴巴自有的基准测试,覆盖 721 个专业领域的工业工作任务。在该基准上,Qwen4-Max 完成相同任务所用的 Token 比上一代少约四分之一到三分之一。在智能体基准测试 Terminal-Bench 4.0 上,Token 用量下降近一半。

据阿里巴巴称,Work360 每项任务的平均成本下降 62%,从 4.43 美元降至 1.67 美元。同期基准测试得分从 47.74 升至 52.69。

Qwen4-Max 在两项基准测试中都胜过上一代,同时 Token 用量远低于后者。 | 图片:阿里巴巴

在外部基准测试中,结果喜忧参半。在 Agents' Last Exam v1.1 这类长周期智能体任务上,Qwen4-Max 得 46.7 分,是对比组中的最佳成绩。在 Humanity's Last Exam 上,它落后于 Opus 5.5 和 GPT 系列模型。在 HealthBench Professional 上,它落后于 GPT 系列模型,但略胜 Opus 5.5。

阿里巴巴称,Qwen4-Max 在智能体和办公任务上领先,而 Opus 5.5 和 GPT 系列模型在 Humanity's Last Exam 上更胜一筹。| 图片:阿里巴巴

在成本仅为前代三分之一的情况下,Qwen4-Max 在中国模型中排名第二

在最高推理档位下,Qwen4-Max 在当前 Artificial Analysis Intelligence Index 上得 43 分。这比 Qwen3.8-Max 高 9 分,比排名中国模型第一、得分 46 分的 Kimi K3.1 Pro 低 3 分。相比前代提升最大的项目是 HLE-Diamond(提升 21 分至 45%)、AA-Omniscience(提升 18 分)和 AutomationBench-AA(提升 17 分)。

Qwen4-Max 在 Artificial Analysis Intelligence Index 上提升 9 分,而每个任务的成本仅为前代的三分之一左右。| 图片:Artificial Analysis

Artificial Analysis 估算,Qwen4-Max 每个 Intelligence Index 任务的成本为 3.27 美元。这约为 Qwen3.8-Max 的 10.02 美元的三分之一,比 Kimi K3.1 Pro 的 3.68 美元低约 11%。节省来自三个方面:Token 价格比前代低 25%,缓存命中成本低 70%,而且该模型消耗的 Token 更少。Qwen4-Max 每个基准测试任务需要约 105,000 个输出 Token,而 Qwen3.8-Max 为 171,000 个。

腾讯的 Hy4 便宜得多,每任务 1.03 美元,但在该指数上低 8 分。在 Terminal-Bench 4.0 上,Qwen4-Max 得分为 48%,落后于 Kimi K3.1 Pro 的 67%。在 Cognition 的编程基准测试 FrontierCode 2.0 上,它与 Kimi 并列 26%,落后于腾讯 Hy4 的 31%。

通过自有 API,阿里巴巴对每百万输入 Token 收费 1.50 美元,每百万输出 Token 收费 4.50 美元。缓存命中费用为 0.075 美元。

Live Delivery 将长时间运行智能体的控制权交还给用户

与 AI 行业中的许多公司一样,阿里巴巴此次发布围绕能够自主处理长任务的智能体来构建。如果用户只在最后才看到结果,就几乎无法纠正错误。阿里巴巴的答案是「Live Delivery」。模型会尽早交付一个可用的初版,并在后续迭代中逐步完善。用户可以在任何时候介入并改变方向。阿里巴巴已开源配套的 Qwen-Live-Delivery-Plugin。

阿里巴巴列举了若干合作伙伴的部署案例。汽车行业合作伙伴 SANF 和 Pacific Union 据称使用 Qwen4-Max,将单个工件的手工编程时间从约 475 分钟缩短至 24 分钟。在金融领域,该模型调用 90 个 MCP 工具从多个来源获取市场数据。它将结果以可验证的 Excel 和 PowerPoint 文件形式交付,每一个数字都可追溯到其原始来源。

阿里巴巴还在内部使用该模型来更高效地运行自身的云基础设施。据该公司称,Qwen4-Max 与内部求解器 MindOpt 配合,将数据标注算力任务的分配成本平均降低 10.42%。在数据中心分配 CPU 任务时,它找到最优解的速度比对照方法快 14 倍以上。

阿里巴巴的模型现在可以自行运行实验室实验

阿里还展示了 Qwen4 代模型在实体实验室中的实际运作。据阿里介绍,该模型与初创公司 Deep Principle 的系统 Mira 配合,自主运行了 48 小时。在此期间,它设计、合成并测试了一种用于高性能数据中心芯片的新型冷却液。它检索了 270 万个候选分子,操控了一套自动化实验室,并完整完成了四轮设计、合成与测量的循环。

Qwen4-Max 还与 PharmMaster 展开了协作,后者是总部位于北京的 AI 药物研发初创公司 DP Technology 旗下的平台。目标是针对一种与阿尔茨海默病相关的炎症反应所设计的化合物。经过多轮文献调研、设计与计算,该模型优化了 24 个候选分子。据阿里称,最终候选分子与靶标蛋白的结合能力,比专利中已知的起始化合物高出约 3.6 倍。DP Technology 的专家认为该结果颇具前景,值得推进后续实验室测试。

Qwen4-Max 的目标是从单条提示词构建出可玩的游戏

阿里还展示了 Qwen4-Max 与 Unity、Unreal Engine、Godot 等成熟游戏引擎的协作。仅凭一段自然语言描述,该模型就应能完成整个开发流程,涵盖场景搭建、游戏逻辑以及外部素材匹配,还包括材质、UI 和视觉效果,直至游戏可玩。

Qwen4-Max 在 Unreal Engine 中搭建了一款巫师冒险游戏,玩家扮演一只猫,与一波波敌人和 Boss 战斗。|图片:阿里巴巴

在开发过程中,该模型自行运行游戏,并利用视觉和交互反馈来调整镜头运用、角色动画、受击反应和光照。阿里巴巴的示例项目是“Meow of Magic”,这是一款 Unreal Engine 冒险游戏,讲述一位猫巫师,玩家在游戏中施放法术并迎战一波波敌人。据阿里巴巴称,Qwen4-Max 还能在 Blender 中构建 3D 模型,这些模型可复用于游戏和 3D 打印。

Anthropic 最近将其编码助手 Claude Code 推向了相反方向。在一波非官方的 Halo 和 GTA: Vice City 克隆作品之后,它禁止该工具复刻现有游戏。

阿里巴巴称,该模型编写更安全的代码,并且不会带来实际的 CBRN 能力提升

在安全基准测试 Last Secure Code Bench 中,据阿里巴巴称,既可用又安全的代码占比从 22% 升至 31.8%。在全部三个测试级别中均被评为安全的代码占比几乎翻倍,从 8% 升至 14.7%。阿里巴巴还测试了 CBRN 风险(化学、生物、放射性与核)。该公司表示,内部评审人员检查了该模型的完整思维链。他们未发现任何真正能够使用户能力显著提升的创新能力。

原文较长,译文为节选;译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Alibaba's Qwen4-Max takes on Opus 5.5, GPT-6, and Kimi K3.1 Pro at much lower prices)

Alibaba launched Qwen4-Max, whose revamped architecture processes long texts more efficiently and with less compute.

On the Artificial Analysis Intelligence Index, the model ranks second behind Kimi K3.1 Pro but runs at roughly a third of its predecessor's cost per task.

A feature called "Live Delivery" lets users inspect intermediate results from autonomous agents early and adjust running tasks on the fly.

Qwen4-Max is Alibaba's answer to Opus 5.5, GPT-6, and Kimi K3.1 Pro. The new flagship pairs a more efficient architecture with much lower prices and targets professional work in industry, science, and finance.

The mixture-of-experts model has 2.4 trillion parameters but activates only 71 billion per token, according to the Qwen team. It processes images natively and offers a one-million-token context window.

A hybrid architecture makes long contexts cheaper to process

Qwen4-Max debuts the revamped architecture of the Qwen4 series. Instead of relying only on standard attention layers, Alibaba combines two methods. Three layers compress the prior context into a condensed state, and a fourth layer then picks out only the parts that matter. That sharply cuts compute costs on long inputs, since the model no longer has to give every part of the context equal attention.

Architecture diagram of Qwen4-Max showing three Gated DeltaNet layers for every Qwen Sparse Attention layer, gated residual read and write steps, MoE blocks, and an N-gram embedding layer at layer 2.Qwen4-Max compresses prior text across three layers before a fourth layer takes a closer look at only the important sections. | Image: Alibaba

The architecture also adds four parallel connections between model layers, with dynamic gates meant to stabilize the flow of information. An N-gram embedding layer with 328 billion parameters supplies common token sequences as ready-made building blocks. According to Alibaba, training the 2.4-trillion-parameter configuration ran stably. Training loss fell steadily throughout the run with no sudden spikes.

Line chart of Qwen4-Max's training loss across the full pretraining run, falling from 1.8 to just above 1.0.Training loss declined steadily throughout the run with no loss spikes, according to Alibaba. | Image: Alibaba

Alibaba's internal evaluations measure model quality in bits per byte across knowledge, code, question answering, and STEM. Qwen4-Max-Base scored best among the five checkpoints tested. It improved 5 to 13 percent over Qwen3.8-Max-Base while activating 25 percent fewer parameters per token.

Four bar charts compare bits per byte for Qwen4-Max-Base, Qwen3.8-Max-Base, GLM-5.3, DeepSeek-V4-Pro-Base, and Kimi K3.1 Pro across knowledge, code, question answering, and STEM.Qwen4-Max-Base posts the lowest, and therefore best, values in all four categories. | Image: Alibaba

Lower token use cuts the cost of each task by more than half

The efficiency gains show up in operating costs. Work360 is Alibaba's own benchmark for industrial work tasks across 721 specialized fields. On it, Qwen4-Max uses roughly a quarter to a third fewer tokens than its predecessor to solve the same tasks. On the agent benchmark Terminal-Bench 4.0, token use drops by nearly half.

According to Alibaba, the average cost per Work360 task falls 62 percent, from $4.43 to $1.67. The benchmark score climbs from 47.74 to 52.69 at the same time.

Two charts plot score against tokens used per task on Work360 and Terminal-Bench 4.0 for Qwen4-Max and Qwen3.8-Max at the Low, High, and Xhigh reasoning levels.Qwen4-Max beats its predecessor on both benchmarks while using far fewer tokens. | Image: Alibaba

Results on external benchmarks are mixed. On long-horizon agent tasks like Agents' Last Exam v1.1, Qwen4-Max scores 46.7 points, the best result in the comparison group. It trails Opus 5.5 and the GPT models on Humanity's Last Exam. On HealthBench Professional, it falls behind the GPT models but edges past Opus 5.5.

Bar charts compare Qwen4-Max with Qwen3.8-Max, Claude Opus 5.5, GPT-6.1 Sol, GPT-6 Astra, and Kimi K3.1 across 16 benchmarks, including Terminal-Bench 4.0, Agents' Last Exam v1.1, HealthBench Professional, and Humanity's Last Exam.Alibaba says Qwen4-Max leads on agent and work tasks, while Opus 5.5 and the GPT models come out ahead on Humanity's Last Exam. | Image: Alibaba

Qwen4-Max ranks second among Chinese models at a third of its predecessor's cost

At its highest reasoning setting, Qwen4-Max scores 43 points on the current Artificial Analysis Intelligence Index. That's nine more than Qwen3.8-Max and three behind Kimi K3.1 Pro, the top Chinese model at 46 points. The biggest gains over the predecessor come on HLE-Diamond (up 21 points to 45 percent), AA-Omniscience (up 18), and AutomationBench-AA (up 17).

This image combines two charts from Artificial Analysis. A bar chart of Intelligence Index v5.0.1 shows Claude Fable 5.2 on top with 65 points and Qwen4-Max at 43 points, with an arrow marking the jump from Qwen3.8-Max at 34 points. A scatter plot compares index scores with cost per task and places Qwen4-Max at about $3, well to the left of Qwen3.8-Max at about $10.Qwen4-Max gains nine points on the Artificial Analysis Intelligence Index and costs only about a third as much per task as its predecessor. | Image: Artificial Analysis

Artificial Analysis puts the cost of Qwen4-Max at $3.27 per Intelligence Index task. That's about a third of Qwen3.8-Max's $10.02 and roughly 11 percent less than Kimi K3.1 Pro's $3.68. The savings come from three places. Token prices are 25 percent lower than the predecessor's, cache hits cost 70 percent less, and the model burns through fewer tokens. Qwen4-Max needs about 105,000 output tokens per benchmark task, compared with 171,000 for Qwen3.8-Max.

Tencent's Hy4 is much cheaper at $1.03 per task but scores eight points lower on the index. On Terminal-Bench 4.0, Qwen4-Max scores 48 percent, behind Kimi K3.1 Pro at 67 percent. On Cognition's coding benchmark FrontierCode 2.0, it ties Kimi at 26 percent and trails Tencent Hy4 at 31 percent.

Through its own API, Alibaba charges $1.50 per million input tokens and $4.50 per million output tokens. Cache hits cost $0.075.

Live Delivery hands control of long-running agents back to users

Like much of the AI industry, Alibaba builds this release around agents that handle long tasks on their own. If users only see results at the end, they have almost no way to fix mistakes. Alibaba's answer is "Live Delivery." The model delivers a usable first version early and builds it out in later passes. Users can step in and change course at any point. Alibaba has open-sourced the accompanying Qwen-Live-Delivery-Plugin.

Alibaba cites several partner deployments. Automotive partners SANF and Pacific Union reportedly used Qwen4-Max to cut manual programming time for a single workpiece from about 475 minutes to 24. In finance, the model taps 90 MCP tools to pull market data from multiple sources. It delivers the results as verifiable Excel and PowerPoint files, and every number traces back to its original source.

Alibaba also uses the model internally to run its own cloud infrastructure more efficiently. According to the company, Qwen4-Max worked with the in-house solver MindOpt to cut the cost of allocating compute jobs for data annotation by an average of 10.42 percent. When distributing CPU tasks in data centers, it found optimal solutions more than 14 times faster than the comparison method.

Alibaba's models now run lab experiments on their own

Alibaba also shows the Qwen4 generation at work in physical labs. Paired with Mira, a system from startup Deep Principle, the model ran autonomously for 48 hours, according to Alibaba. In that time, it designed, synthesized, and tested a new coolant for high-performance data center chips. It searched 2.7 million molecular candidates, controlled an automated lab, and completed four full cycles of design, synthesis, and measurement.

Qwen4-Max also worked with PharmMaster, a platform from Beijing-based AI drug discovery startup DP Technology. The target was a compound aimed at an inflammatory response linked to Alzheimer's disease. Over several rounds of literature review, design, and computation, the model optimized 24 candidate molecules. According to Alibaba, the final candidate binds to the target protein about 3.6 times better than the known starting compound from a patent. DP Technology's experts consider the result promising enough to pursue further lab tests.

Qwen4-Max aims to build playable games from a single prompt

Alibaba also shows Qwen4-Max working with established game engines like Unity, Unreal Engine, and Godot. From a single natural-language description, the model is supposed to handle the entire development process. That covers scene setup, game logic, and matching external assets, along with materials, UI, and visual effects, until the game is playable.

Scene from the game "Meow of Magic" showing a cat in a wizard's robe and hat inside a medieval hall with stained-glass windows.Qwen4-Max built a wizard adventure in Unreal Engine in which players, as a cat, battle waves of enemies and bosses. | Image: Alibaba

During development, the model runs the game itself and uses visual and interactive feedback to tweak camera work, character animations, hit reactions, and lighting. Alibaba's sample project is "Meow of Magic," an Unreal Engine adventure about a cat wizard in which players cast spells and fight waves of enemies. According to Alibaba, Qwen4-Max also builds 3D models in Blender that can be reused for games and 3D printing.

Anthropic recently moved its coding assistant Claude Code in the opposite direction. After a wave of unofficial Halo and GTA: Vice City clones, it barred the tool from recreating existing games.

Alibaba says the model writes safer code and offers no real CBRN uplift

On the safety benchmark Last Secure Code Bench, the share of code that is both functional and secure rose from 22 to 31.8 percent, according to Alibaba. The share rated secure at all three testing levels nearly doubled, from 8 to 14.7 percent. Alibaba also tested for CBRN risks (chemical, biological, radiological, and nuclear). Internal reviewers went through the model's full chains of thought, the company says. They found no real capacity for innovation that would give a user a significant boost in capability.

Qwen4-Max is available through QwenCloud and in QwenWork, Qoder, and QwenStudio. Alibaba plans to release smaller Qwen4 models as open source later.

In August, Alibaba released Qwen3.8-Max, its first Max model built for autonomous tasks that run for days. It also has 2.4 trillion parameters, but 95 billion are active per token. At the time, Artificial Analysis found that the model needed 64 steps per task instead of 14 and cost more than twice as much per task as Qwen3.7-Max.

In late August, Alibaba previewed the Qwen4 architecture with Qwen3.8-Flash-Next, which already used an N-gram embedding layer with 51 billion parameters. The company had released the Qwen3.8-Max weights in mid-August. For Qwen4, only the smaller models will get open weights for now.

Moonshot AI followed in October with Kimi K3.1 Pro. At 2.8 trillion parameters, it's even larger, and Moonshot says it sits just behind Claude Opus 5.5 on Terminal-Bench 4.0. Like Alibaba, DeepSeek uses a hybrid attention design with sparse attention in V4-Pro to process long contexts more cheaply. Zhipu AI took a different route and got GLM-5.3's gains entirely from extra post-training.

AI News Without the Hype – Curated by Humans

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 The Decoder(Alibaba's Qwen4-Max takes on Opus 5.5, GPT-6, and Kimi K3.1 Pro at much lower prices),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗