Alibaba releases Qwen4-Max and claims 62% lower task costs
Original title: Alibaba touts 62% lower task costs with Qwen4-Max, adds cyber safeguards
The 2.4-trillion-parameter model activates about 71 billion parameters per token and supports a 1-million-token context.
Alibaba released Qwen4-Max, keeping the 2.4-trillion-parameter mixture-of-experts design of Qwen3.8-Max but activating about 71 billion parameters per token instead of about 95 billion. On its Work360 benchmark, Alibaba said the model used 35% fewer tokens in High effort mode and cut average task cost from $4.43 to $1.67. New list prices are $1.50 per million input tokens and $4.50 per million output tokens. Alibaba also said training-data filters keep defensive cybersecurity material and remove offensive attack-enhancement data, raising Last Secure Code Bench pass rates from 22% to 31.8%. Analysts said buyers should verify both the savings and the safeguard claims independently.
Unconfirmed: The 62% cost cut and security gains are Alibaba’s figures; analysts say they do not assure a buyer’s spending or that offensive capability was removed.
- Alibaba cited Code Arena, Agents’ Last Exam and a programming-time demo falling from about 475 minutes to 24.
- Smaller Qwen4 models with open weights are planned; commentators warned fine-tuning can remove safeguards.
- An October Cohere report concerned Qwen3.8-Flash-Next, not Qwen4-Max.
- Prompt retention of up to 30 days, and longer for flagged content, was flagged for non-China deployments.
- Vendors
- Alibaba
- Organizations
- Greyhound ResearchSecurity BrigadeCohereCheck Point
- AI models
- Qwen4-MaxQwen3.8-Max
From the original article
4 min read at csoonline.com
Alibaba has released Qwen4-Max, a 2. 4-trillion-parameter flagship model, claiming it reduces the average cost of completing professional tasks by 62% compared with its predecessor while introducing additional safeguards against offensive cybersecurity capabilities and other high-risk uses.
csoonline.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.