AIO Apex
en

Alibaba launches Qwen4-Max with 2.4-trillion-parameter MoE and million-token context

Alibaba Cloud
Share:
Alibaba launches Qwen4-Max with 2.4-trillion-parameter MoE and million-token context

Alibaba released Qwen4-Max on November 16, its most powerful language model to date, built on a 2.4-trillion-parameter mixture-of-experts backbone and capable of processing up to one million tokens in a single context window. The model accepts text, image, and video input, and is available immediately through the QwenCloud API, Qwen Studio, QwenWork, and Qoder.

The launch marks Alibaba's most direct challenge yet to frontier AI providers. API pricing comes in at $1.50 per million input tokens and $4.50 per million output tokens, with cached input at just $0.075 per million — competitive positioning aimed squarely at enterprise customers who run large-scale agent workloads. Alibaba says the model uses roughly 25% fewer active parameters than its predecessor Qwen3.8-Max, reducing inference costs while maintaining capability gains.

Alibaba's internal Work360 benchmark shows average task costs dropping from $4.43 to $1.67 — a claimed 62% reduction — but this figure combines token usage and pricing in ways that make direct comparison difficult. Independent testing by Artificial Analysis puts Qwen4-Max at $3.27 per Intelligence Index task, versus $3.68 for Kimi K3.1 Pro.

On raw capability, the picture is mixed. Artificial Analysis ranks Qwen4-Max at 43 on its Intelligence Index, well behind Claude Fable 5.2 (65), Gemini 4 Carbon (60), Claude Opus 5.5 (59), and Claude Sonnet 5.5 (55). The model performs competitively on DeepSWE and APEX-Agents benchmarks but trails substantially on Terminal-Bench 4.0, ProgramBench, and ChartoGraphy. Speed is also a concern: Artificial Analysis clocks roughly 32 minutes per Intelligence Index task and about 159 seconds to the first output token.

Alibaba has promised open-weight, smaller-scale Qwen4 models for developers, but has not given a release date or confirmed licensing terms. The API supports tool calling, structured outputs, web search, batch processing, and both OpenAI and Anthropic API formats, making migration straightforward for teams already using those ecosystems.

As reported by VentureBeat, the launch comes amid intensifying cost competition in the enterprise AI market, where every major lab is racing to lower the price-per-useful-task threshold. Qwen4-Max positions Alibaba as a credible lower-cost alternative for workloads where frontier-level accuracy is not critical, though for tasks requiring top-of-chart reasoning, the gap to leading Western models remains significant.

Originally reported by Alibaba Cloud. Read the original article for additional details.

View original source
Share: