Release5 sources438 comments in the community
Alibaba releases Qwen4-Max, its new flagship 2.4T-parameter MoE model with a 1M-token context window, saying it beats Qwen3.8-Max with up to ~50% fewer tokens (Qwen)
Techmeme ·
The Brief · written from 5 sources
- Alibaba launched Qwen4-Max, a 2.4T-parameter MoE model with a 1M-token context window, and says it beats Qwen3.8-Max while using up to ~50% fewer tokens.
- Alibaba says Qwen4-Max beats Fable 5.1 and Kimi K3.1 Pro in coding benchmarks. The model is not open weight, but Alibaba plans to release weights for smaller Qwen4 models.
- Artificial Analysis scores Qwen4-Max 43 on its Intelligence Index, 9 points above Qwen3.8 Max, with about 67% lower cost per task than Qwen3.8 Max.
Why it matters
The flagship is not open weight, so developers can only use it through Qwen's hosted access for now, and independent benchmark results are mixed. Test it on your own workloads before switching.
Who’s affected
- Developers using hosted Chinese AI models
- Teams waiting for open-weight Qwen4 releases
- Qwen4-Max
- Qwen3.8-Max
- Kimi K3.1 Pro
- Fable 5.1
- #ai-models
- #qwen
- #alibaba
- #open-weights
- #llm-benchmarks
Qwen : Alibaba releases Qwen4-Max, its new flagship 2.4T-parameter MoE model with a 1M-token context window, saying it beats Qwen3.8-Max with up to ~50% fewer tokens — Today, we are introducing Qwen4-Max, your new AI teammate for work, creativity, and discovery.
What developers are saying
Based on 438 comments · Mixed reactions
Commenters are split. Some praise the focus on efficiency and are waiting for smaller Flash and 27B versions, while others call the flagship disappointing for its size and its Artificial Analysis score of 43. Concerns also include slow output speed, the lack of open weights, and censored answers on Chinese political topics.
- Underwhelming benchmark results30%
- Waiting for smaller models25%
- Censorship on political questions15%
- Speed and hosting access15%
- Praise for efficiency15%
In favour
- + Strong focus on efficiency, as praised in comments
- + Smaller tiers such as Flash and 27B are seen as strong by some commenters
- + Close to the frontier on some traditional benchmarks, according to one commenter
Concerns
- − Artificial Analysis score of 43 seen as underwhelming
- − Output speed reported as slow, around 30 tokens per second for Qwen 3.8
- − Not open weight, and no clear API access path yet
- − Concerns about censored answers on political questions
- − Some worry about hidden behaviours in open weights
“This is exciting. The continued focus on efficiency alongside quality is encouraging”
“Disappointing performance for a 2.4T model in the end of 2026.”
“Only 43 on AA, no where near the frontier now, looks like distillation blocking has halted their progress.”
Timeline
Full coverage · 5 sources
- 2d ago ↗
- 2d ago ↗
- 2d ago ↗
Techmeme
Alibaba releases Qwen4-Max, its new flagship 2.4T-parameter MoE model with a 1M-token context window, saying it beats Qwen3.8-Max with up to ~50% fewer tokens (Qwen)
- 2d ago ↗
Hacker News: Front Page
Qwen4-Max - 2d ago ↗
Bloomberg Technology
Alibaba Unveils New Flagship AI Model in Riposte to US Rivals