Skip to content
latestai.eulatestai.eu
Q3

Qwen 3.8 Max

by Alibaba

Try itWebsiteDocsHugging Face

Alibaba's open-weights 2.4T-parameter MoE language model (95B active) from the Qwen3.8 family, with a hosted Max version.

Available in EULanguage ModelsCodingModel

Updated EU status checked Last shown

Qwen 3.8 Max at a glance

What is Qwen 3.8 Max?
Qwen 3.8 Max is a model from Alibaba. Alibaba's open-weights 2.4T-parameter MoE language model (95B active) from the Qwen3.8 family, with a hosted Max version.
Who makes Qwen 3.8 Max?
Qwen 3.8 Max is made by Alibaba.
Does Qwen 3.8 Max work in the EU?
Available in the EU. Weights are downloadable from Hugging Face with no EU exclusion in the provided material, though the custom license text was not seen in full. Hosted access is via Qwen Cloud and Qwen Studio; EU data residency was not confirmed. Last verified 14 Nov 2026.
How much does Qwen 3.8 Max cost?
Open source. Weights are free to download under the custom Qwen3.8-Max license. Hosted Qwen3.8-Max is available through the paid Qwen Cloud API; pricing was not given in the provided material. Free chat access via Qwen Studio. The release date is taken from the Hugging Face repo creation date. Licence: Qwen3.8-Max license (custom).
Where was Qwen 3.8 Max shown?
In 8 creator videos by AI Search, Matthew Berman, Matt Wolfe. Most recently by AI Search on 3 Sept 2026, at 30:12 in “Claude Fable 5.1 is savage”.

Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.

About

Qwen3.8-2.4T-A95B is a causal language model from Alibaba's Qwen team and the basis for the hosted Qwen3.8-Max. It is a mixture-of-experts model with 2.4T total and 95B active parameters, 92 layers, and 512 experts (10 routed plus 1 shared). It uses a hybrid layout of Gated DeltaNet and Gated Attention blocks and was trained with multi-token prediction. Native context is 262,144 tokens, extensible to about 1,010,000.

Alibaba describes it as the first Qwen-Max-class model released as open weights. The focus is coding, professional work, research and long-horizon agentic tasks, with better planning and handling of environment feedback. Reasoning depth is adjustable via reasoning_effort, and preserve_thinking keeps reasoning from earlier messages. Creators cite a score of 58 on the Artificial Analysis index and a second-place Arena WebDev ranking; these are third-party reports, not verified here.

The weights are on Hugging Face in Transformers/safetensors format and work with vLLM, SGLang and similar engines. Creators put the download at roughly 4.8 TB, so self-hosting needs a large multi-GPU setup. The license is a custom "qwen3.8-max" license, not a standard open-source one. Alibaba's Qwen Cloud API offers the official Qwen3.8-Max, which adds vision input, non-thinking mode, 1M context by default and built-in tools. The open checkpoint is text-only.

Key features
  • 2.4T total / 95B active parameter MoE with 512 experts
  • Hybrid Gated DeltaNet and Gated Attention architecture
  • 262K native context, extensible to ~1M tokens
  • Adjustable reasoning effort and preserved thinking across turns
  • Compatible with vLLM, SGLang and Transformers
  • Hosted Qwen3.8-Max adds vision input, non-thinking mode and built-in tools
8 moments

As seen on YouTube

Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.

MA
Matthew Berman4 Aug 2026
Open-source is WINNING

New open-weights ~2.4T parameter model from Alibaba that is competitive with top US closed models on benchmarks (Terminal Bench 86.6 vs 84.6 for Fable), leads in multimodal, document and spatial categories. Highlighted for reproducing research papers with self-generated improvements and autonomous chip design. Priced at $2/M input and $6/M output on OpenRouter, far below GPT-5.6 Soul and Fable.

0:00Open on YouTube
Keep exploring

Similar AI tools

Closest neighbours by meaning, not just by category.

Q3

Qwen 3.5 122B

Model · Alibaba

1

Alibaba's Qwen 3.5 large open-weights language model (~122B parameters) from the Qwen family.

Available in EULanguage Models
Release date n/aOpen source
Q3

Qwen 3.8 Flash Next

Model · Alibaba

2

Alibaba's Qwen3.8 Flash Next: a 125B-parameter MoE language model with about 6B active parameters, previewing Qwen 4 design.

Available in EULanguage Models
Release date n/aOpen source
Q3

Qwen 3.7 Max

Model · Alibaba

2

Alibaba's Qwen flagship large language model, offered as a hosted proprietary model via Qwen Chat and Alibaba Cloud.

UnverifiedLanguage Models
Release date n/aAPI
Q3

Qwen 3.5 0.8B

Model · Alibaba

1

Small 0.8B-parameter open-weights language model in Alibaba's Qwen 3.5 family, aimed at lightweight and on-device use.

Available in EULanguage Models
Release date n/aOpen source
Q2

Qwen 2.5 0.5B

Model · Alibaba

1

Small 0.5B-parameter open-weights base language model from Alibaba's Qwen2.5 series, Apache 2.0 licensed.

Available in EULanguage Models
15 Sept 2024Open source
Q3

Qwen 3.8 27B

Model · Alibaba

3

Compact 27B-parameter model from Alibaba's Qwen family, runnable locally on a single high-end consumer GPU.

Available in EULanguage Models
Release date n/aOpen source