Day 7 receipt
Qwen4-Max
Day 1 of 7Claims on the record
We check on 23 Nov 2026
- They promised
- A flagship model that matches top US models on coding and agent work at about 60% lower cost per task.
The claims we’ll check
- “Qwen4-Max achieves top rankings across leading benchmarks, particularly in coding, agentic tasks and visual reasoning”
- “outperforms its predecessor, Qwen3.8-Max, overall while using 35% fewer tokens at High effort reasoning mode and reducing the average cost per task by around 60%”
- “Open-weight, smaller-scale models from the Qwen4 series will soon be released to developers worldwide”
- “Qwen4-Max activates only 71 billion per token—25% fewer activated parameters compared to Qwen3.8-Max at the same 2.4T scale”
Quoted word for word from Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery. Day 1: the benchmark and cost claims are Alibaba's own, and the model's weights are closed. No independent tests yet.
Full evidence 4 claims · 0 user reports · 4 sources
Alibaba's new flagship Qwen model: a 2.4-trillion-parameter mixture-of-experts with native vision and a one-million-token context window, available through QwenCloud and Alibaba's apps, with closed weights.
Launch post: Alibaba Unveils Qwen4-Max: New AI Teammate for Work, Creativity, and Discovery
Why this stamp
Day 1: the benchmark and cost claims are Alibaba's own, and the model's weights are closed. No independent tests yet.
Alibaba says Qwen4-Max beats its predecessor while using fewer tokens and cutting average task cost by about 60% on its own Work360 benchmark, and that it ranks among the leaders in coding. The Next Web, Quartz and the South China Morning Post confirm the launch, the closed weights and the safety measures, but all repeat Alibaba's numbers. The Next Web also notes mixed results on coding charts compared with Claude Opus 5.5 and Kimi K3.1. No independent tester or user reports exist yet. We re-stamp on day 7 (23 November) and day 14.
Launch dayWhat they claimed, quoted word for word · 4
“Qwen4-Max achieves top rankings across leading benchmarks, particularly in coding, agentic tasks and visual reasoning”
The scores are Alibaba's own. TNW cites charts that show it just ahead of Claude Opus 5.5 on one coding test and well behind on another, so 'top rankings' is only partly backed so far.
“outperforms its predecessor, Qwen3.8-Max, overall while using 35% fewer tokens at High effort reasoning mode and reducing the average cost per task by around 60%”
Measured on Alibaba's own Work360 benchmark. Quartz and Tech in Asia repeat the cost cut, none has tested it independently.
“Open-weight, smaller-scale models from the Qwen4 series will soon be released to developers worldwide”
No date given. TNW and Quartz note Qwen4-Max itself ships without open weights, unlike earlier Qwen releases.
“Qwen4-Max activates only 71 billion per token—25% fewer activated parameters compared to Qwen3.8-Max at the same 2.4T scale”
An architecture claim only the vendor can confirm.
OthersWhat outlets and testers say · 4
Says the scores are Alibaba's own and unverified, cites charts showing mixed coding results against Claude Opus 5.5, and notes the weights are closed.
“Alibaba launches Qwen4-Max flagship AI model without open weights” · Open the source
Reports the 62% task-cost cut on Work360 as Alibaba's claim and that smaller Qwen4 models are planned as open source.
“Alibaba is launching a new flagship AI model, claiming to rival U.S. leaders” · Open the source
Focuses on the cybersecurity guardrails: offensive attack data filtered out of training, defensive material kept.
“Alibaba says new flagship AI model cuts task costs by 60% while tightening cyber guardrails” · Open the source
Confirms the parameter count, one-million-token context and the plan to open-source smaller Qwen4 models over time.
“Alibaba launches Qwen4-Max AI model, cutting task costs 60%” · Open the source