Qwen4-Max Becomes Top Chinese Model on Agent Arena

Alibaba's new Qwen4-Max debuts as the highest-ranked Chinese model on Arena's agent leaderboard, outscoring Tencent, Moonshot and DeepSeek.

@sharedot

Qwen4-Max Becomes Top Chinese Model on Agent Arena

By @sharedot · · 6 pages

  • AI Frontier
  • Open Weight Models
  • Benchmarks
  • Alibaba Qwen

Alibaba's new Qwen4-Max debuts as the highest-ranked Chinese model on Arena's agent leaderboard, outscoring Tencent, Moonshot and DeepSeek.

A Chinese Model Breaks Into Arena's Agent Rankings

Alibaba released Qwen4-Max on Monday, November 16, calling it the most capable version yet of its Qwen family, with 2.4 trillion total parameters, native vision capabilities and a one-million-token context window.

The striking result for practitioners is where it landed on Arena's agent leaderboard: according to KuCoin, citing CoinDesk, Qwen4-Max is the highest-ranked Chinese model on the board, placing Alibaba ahead of Tencent, Moonshot and DeepSeek in that evaluation.

The model scored 7.46% across 14,727 sessions and ranked 14th among single models, helping Alibaba reach fifth place in the overall lab ranking.

Why the Upset Matters for the Frontier Race

The surprise is that this is a consensus-breaking result, not a scheduled benchmark drop.

PYMNTS reports Alibaba's own release claims top rankings on leading benchmarks, particularly in coding, agentic tasks and visual reasoning, and says Qwen4-Max outperformed rivals including Anthropic's Claude Fable 5.1, Moonshot's Kimi K3.1 Pro and the Muse Wonder Full model.

Yet the nuance matters: KuCoin reports that U.S. labs still dominate the overall agent and web-development rankings on Arena, with Qwen4-Max only closing the gap.

For builders tracking open-weight ecosystems, the competitive picture is no longer a two-horse race — Moonshot's Kimi K3 has pushed open-weight models near Western standards while DeepSeek competes hard on API pricing.

The Evidence: Coding, Vision and Cost

The technical evidence is unusually concrete.

On Code Arena — a real-world coding leaderboard decided by one-on-one user voting — Qwen4-Max debuted sixth globally with 1,750 points, a gain of 74 points over its predecessor, and ranked eighth in both brand design and reference-image-based design categories, according to KuCoin.

PYMNTS independently cites the same sixth-place Code Arena ranking.

Alibaba's proprietary Work360 benchmark, which spans 721 specialized fields, shows Qwen4-Max outperforming Qwen3.8-Max while using 35% fewer tokens in high-effort reasoning mode and cutting average cost per task by roughly 60%, PYMNTS reports.

KuCoin adds that through QwenCloud the model is priced at $1.50 per million input tokens and $4.50 per million generated tokens, a 25% reduction from the previous flagship.

One Catch: The Flagship Stays Closed

Here is the reversal for the open-weight community: Qwen4-Max ships with proprietary weights.

KuCoin reports developers can use it via API or Alibaba's apps but cannot download, modify or run it on their own hardware — even though Qwen remains the world's most-downloaded open-weight model family, with 151,448 derivative models on Hugging Face, roughly 4.7 times the number built on Meta's Llama family.

Alibaba says smaller Qwen4 models will be released with open weights, but has not disclosed their sizes or dates.

The backdrop is an aggressive company-wide push: PYMNTS, citing Bloomberg News, reports CEO Eddie Wu said Alibaba's T-Head chipmaking division, which filed to go public in Hong Kong, would introduce China's most powerful AI chips domestically, and that Alibaba has made achieving AGI its primary goal.

Sources

  1. pymnts.com › Alibaba Takes on Anthropic With New Qwen AI Model | PYMNTS.com
  2. kucoin.com › Alibaba Launches Qwen 4 Max, Tops Chinese AI Agent Rankings | KuCoin

More on AI Frontier