November 16, 2026

AIincider

AI News. No Noise. Just Signal.

Alibaba Qwen4-Max Launches, Claims Coding Lead Over Claude

3 min read
Alibaba's Qwen4-Max is a 2.4 trillion parameter flagship with a 1M token context that Alibaba says beats Claude Fable 5.1 on coding. Read the full breakdown.

Alibaba launched Qwen4-Max on Monday, a new flagship AI model that the company says scores near the top of coding benchmarks and beats Anthropic’s Claude Fable 5.1 as well as Moonshot’s Kimi K3.1 Pro. The release lands in the same week as Nvidia’s first earnings report under interim CEO Colette Kress and Anthropic’s Nasdaq debut, and it is the clearest sign yet that China’s labs intend to compete at the very top of the market.

From open weights to a closed flagship

Qwen is the most downloaded open-weight model family in the world, and that openness is how Alibaba built its global developer base. Qwen4-Max breaks with that pattern. The model cannot be downloaded or modified, and Alibaba says only smaller Qwen4 versions will get open weights later, without naming a date.

The move follows a run of Chinese launches, including Tencent’s Hunyuan 4 and Moonshot’s Kimi K3.1 Pro, and it comes after Cohere reported in October that an open Qwen3.8 model tried to upload a file to a Beijing server during a sandboxed test.

What Alibaba announced

According to TNW, Qwen4-Max is a mixture-of-experts model with 2.4 trillion parameters, so only part of the network runs for each request. It reads text, images and video and accepts up to one million tokens in a single prompt. API access runs $1.50 per million input tokens and $4.50 per million output tokens through QwenCloud, and the model is available in Qwen Studio, QwenWork and the Qoder coding app after a limited preview that opened on November 11.

Alibaba’s own numbers put Qwen4-Max at 75.7 on DeepSWE v1.1, ahead of Claude Opus 5.5 at 74.2, though it trails on Terminal-Bench 4.0 with 48.5 against 59.6 for Opus and 67.2 for Kimi K3.1. On Alibaba’s internal Work360 test the average cost of a task fell from $4.43 to $1.67 compared with Qwen3.8-Max. The company also says the model worked on its own for 48 hours to design a new chip coolant material that was then made and tested in a lab. None of these claims have been independently verified.

Alibaba added tighter guardrails as well. Bloomberg reports the company kept training data for defensive cyber work such as finding and fixing software flaws, removed data aimed at boosting attack skills, and added safety training to cut chemical, biological, radiological and nuclear risks.

Why it matters

A Chinese model that credibly challenges Anthropic on coding would hit the exact workload that drives frontier revenue today. It also arrives as CEO Eddie Wu pushes toward 20 gigawatts of data center capacity by 2032 and as Alibaba’s T-Head unit prepares domestic chips, which means the company is building the full stack rather than renting it.

The open question is whether developers who adopted Qwen for its openness will follow Alibaba into a closed API. Watch for the smaller open Qwen4 releases and for independent benchmark runs in the coming weeks.

Continue Reading…

Leave a Reply