Alibaba Opens Qwen3.8-Max Weights, Letting Teams Self-Host a 2.4T Model
Alibaba drops open weights for Qwen3.8-27B and the massive 2.4T-A95B Max model, both free under Apache 2.0
- Qwen3.8-27B open weights are live on Hugging Face under Apache 2.0, a native multimodal dense model with 262K context.
- Qwen3.8-2.4T-A95B (Max) weights also released — the first time Alibaba has open-sourced a Max-class model, also Apache 2.0.
- Qwen3.8-27B beats Qwen3.7-Plus (an API-only model) on agentic coding (SWE-bench Pro: 61.7 vs 57.6) and office tasks (CoWorkBench: 70.7 vs 65.1).
- Hybrid Gated DeltaNet architecture: 3 out of 4 attention sublayers use linear O(n) attention, enabling 262K native context without quadratic memory costs.
- Thinking mode is on by default with tunable reasoning effort (xhigh/medium/low) and a
preserve_thinkingflag for multi-turn agent coherence. - Max model API pricing: $2/$6 per million input/output tokens on QwenCloud; 27B hosted API coming soon.
Alibaba just made good on a promise. Qwen3.8 open weights are now live on Hugging Face and ModelScope, delivering two very different models for very different hardware budgets: the compact Qwen3.8-27B, a deployment-friendly dense model, and the flagship Qwen3.8-2.4T-A95B, a massive mixture-of-experts (MoE) model that is the first Max-class Qwen release ever made downloadable. Both ship under Apache 2.0.
Two models, one announcement
Alibaba released Qwen3.8-Max as a 2.4-trillion-parameter sparse MoE model with 95 billion active parameters per token. MoE means the model has a huge number of parameters total, but only a fraction of them activate on any given token, keeping inference costs manageable relative to the raw parameter count. This is the first time Qwen is opening the weights of a Max-class model.
Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability.
The architecture under the hood
Qwen3.8-27B is a dense model, meaning every parameter activates on every token. That makes VRAM usage predictable and latency low, which is exactly what you want for local deployment. But fitting a 262K-token context into a dense 27B model without blowing up memory requires a clever architectural trick.
Qwen3.8-27B is organized as 64 layers, structured as 16 repeating blocks, each composed of three Gated DeltaNet sublayers followed by a single Gated Attention sublayer, with feed-forward networks interleaved throughout. Three-quarters of the model's attention budget runs on linear attention; only every fourth sublayer performs full self-attention. Gated DeltaNet is a form of linear attention: instead of computing relationships between every pair of tokens (which scales as O(n²) and becomes ruinously expensive at 262K tokens), it trades a small amount of expressive capacity for O(n) complexity. Full Gated Attention appears only where it earns its keep , roughly once per block , to preserve the long-range precision that linear attention alone cannot.