Alibaba's Qwen4-Max Cuts Task Costs 62% to $1.67
By @sharedot · · 8 pages
- AI Frontier
- Qwen4 Max
- Open Weight Models
- Coding Benchmarks
Alibaba launched Qwen4-Max on November 16, claiming coding wins over Claude and Kimi while slashing average task cost by a reported 62%.
What happened: Qwen4-Max goes live
Alibaba released Qwen4-Max on November 16, 2026, the flagship of its new Qwen4 series, available via the QwenCloud API and Token Plan after a preview that opened November 11.
The multimodal model handles text, images and video, supports a 1 million-token context window, and runs on a sparse mixture-of-experts architecture built around a 2.4-trillion-parameter backbone — meaning each request activates only a subset of its parameters rather than firing up all 2.4 trillion at once.
Alibaba positions it for professional work spanning coding, 3D modelling, game development, finance, law, healthcare and frontier scientific discovery, with demos using Blender and game engines like Unity and Unreal Engine.
Why the efficiency story is the surprise
The surprising part is not another big model — it is the cost curve. According to Business Upturn, Alibaba's Work360 evaluation put average task cost at $1.67, down from $4.43 for Qwen3.8-Max, a reported 62% reduction.
TradingView adds that Qwen4-Max uses roughly 50% fewer tokens than its predecessor, with pricing set at $1.50 per million input tokens and $4.50 per million output tokens, and a cost per task around $3.27.
Because token usage and per-token pricing stack, Alibaba's pitch leans away from the pure capability arms race toward efficiency and practical deployment — a notable reversal from the usual flagship-launch framing.
The benchmark evidence — and the gap
Alibaba says Qwen4-Max scored near the top of coding benchmarks, outperforming Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro, per Bloomberg and Reuters reporting carried by The Business Times and The Straits Times.
TradingView reports the model scored 75.7 on the DeepSWE v1.1 coding benchmark, ahead of Claude Opus 5.5.
But independent rankings tell a narrower story: according to TradingView, they place Qwen4-Max behind competitors such as Kimi on outright performance, while still crediting its favorable cost per task.
Developers evaluating the model should weigh third-party results alongside vendor numbers, especially since coding performance varies by task.
Closed flagship, open family, bigger ecosystem
In a break from the strategy that made Qwen the world's most downloaded open-weight model family, Alibaba kept Qwen4-Max closed, though it says smaller versions of the Qwen4 series will be released soon without a date, per Business Upturn.
That protects the flagship while widening the developer base.
The launch also fits a broader build-out: CEO Eddie Wu has made AGI Alibaba's primary goal, targets 20 gigawatts of global data centre capacity by 2032, and said in September that the T-Head chip unit — which filed for a Hong Kong IPO — would roll out China's most powerful AI chips domestically.
Safety measures arrive with the model
Alibaba introduced upgraded security measures alongside the release, following incidents that raised concern over cybersecurity risks of open-weight models — a move The Straits Times notes follows similar steps by Tencent and Moonshot.
According to Business Upturn, the company says security checks were added to coding training, and material intended to improve offensive hacking was removed from training while keeping vulnerability-finding material; safeguards against misuse for cybercrime and hazardous materials were strengthened, and internal testing found no evidence of significant high-risk capabilities.
These are the developer's own descriptions, not independent proof, but Chinese developers have visibly increased safety disclosures amid global concern over loss-of-control risks.
What comes next for builders
For developers, the practical hook is compatibility: TradingView reports Qwen4-Max works with protocols established by OpenAI and Anthropic, so switching is closer to a configuration change than a rewrite.
Access runs through the QwenCloud API plus QwenWork, Qoder and QwenStudio, and a Token Plan preview window runs until November 25, per Business Upturn.
The open questions are whether the 62% cost claim holds in real deployments outside Alibaba's own evaluation, and whether the promised smaller open-weight Qwen4 models arrive quickly enough to keep the family's download crown.
Watch whether efficiency, not raw benchmark supremacy, becomes the new basis of competition in the flagship tier.
Sources
- businesstimes.com.sg › Alibaba unveils new flagship AI model in challenge to US rivals
- tradingview.com › Alibaba unveils Qwen4-Max, claims it beats rival models on coding
- businessupturn.com › Alibaba launches Qwen4-Max, claims 62% lower cost per task
- straitstimes.com › Alibaba unveils new flagship AI model in riposte to US rivals