Alibaba's Qwen4-Max Tops Coding Benchmarks, Edging Claude
By @sharedot · · 7 pages
- AI Frontier
- Alibaba
- Open Weight Models
- AI Benchmarks
- Qwen4 Max
Alibaba released Qwen4-Max on Nov 16, claiming coding scores that beat Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro.
What happened: the full Qwen4-Max lands
Alibaba launched its new flagship Qwen4-Max model on November 16, 2026, following a preview on November 11.
The Hangzhou company says the model scored near the top of coding benchmarks, outperforming Anthropic's Claude Fable 5.1 and homegrown rival Moonshot's Kimi K3.1 Pro, and claims performance that rivals Anthropic and OpenAI.
Unlike most of the Qwen family, the flagship is not open weight, though Alibaba says smaller versions of the Qwen4 series will be released soon.
Why builders will care: cost per task drops
The surprising number for practitioners is economics, not just benchmark rank.
According to Business Upturn, Alibaba's own Work360 evaluation puts the average task cost at $1.67, down from $4.43 for Qwen3.8-Max — a reported 62% reduction, though that is a vendor figure on the company's own evaluation, not a guarantee for every workload.
Business Upturn also reports a one-million-token context window, text, image and video input support, web search, tool calling, and structured JSON output, with the model pitched for coding, design, and multi-step agentic projects such as Blender 3D work and games built in Unity, Unreal Engine, and Godot.
The evidence and its limits
The headline claims come from Alibaba's own release: near-top coding scores against Claude Fable 5.1 and Kimi K3.1 Pro, and improvement over August's Qwen3.8-Max with fewer tokens.
The Straits Times and The Business Times both report the benchmark framing, but the scores are self-reported rather than independently verified.
Business Upturn similarly cautions that Alibaba's safety assessment relies on internal testing and expert review, and that real spending will depend on the task, settings, and how often work must be repeated — a useful checklist for anyone reproducing the numbers.
Safety gets upgraded too
Alongside performance, Alibaba introduced upgraded security measures after a series of alarming incidents raised concerns about cybersecurity risks of open-weight models, following similar moves from Tencent and Moonshot.
The company says security checks were added to training alongside coding tests to reduce vulnerabilities such as code injection, memory-safety flaws, and broken access controls, with safeguards strengthened against misuse in cybercrime and hazardous materials, and internal testing finding no significant high-risk capabilities.
Chinese developers have increased safety disclosures amid global concern over rogue-AI and loss-of-control risks.
Stakes: the AGI push behind the release
The release fits a fast-moving Chinese push against US leaders, arriving amid Tencent's Hunyuan 4 and Moonshot's Kimi K3.1 Pro. The Qwen series remains the most downloaded open-weight model family in the world under CEO Eddie Wu, who has made artificial general intelligence Alibaba's primary goal.
Alibaba is building a full ecosystem from models to AI chips: Wu said in September that the T-Head chip unit, which recently filed for a Hong Kong IPO, would roll out China's most powerful AI chips domestically, and he targets 20 gigawatts of global data-centre capacity by 2032.