Qwen3.8 Max
Current version: 3.8 Max updated
Alibaba flagship: 2.4T params, 1M context, multimodal
What's new in Qwen3.8 Max
- Context extended to 1,000,000 tokens with output up to 131,072
- 2.4T MoE parameters versus 1T in the previous generation, about 95B active per request
- Native multimodality: text, image and video input
- Terminal-Bench 2.1 score of 86.6 against 51.5 for Qwen3.6 on version 2.0
- SWE-bench Pro at 67.7 versus 49.5 for Qwen3.6
What it's the best tool for
- 1,000,000-token context fits an entire codebase or a full book
- 2.4T MoE parameters with 95B active: flagship quality without flagship pricing
- Multimodal input handling text, images and video in a single request
- Agentic workflows with autonomous runs lasting more than ten days
- Terminal-Bench 2.1 at 86.6 and SWE-bench Pro at 67.7, matching top-tier coding models
- Strong multilingual output that does not read like a translation
When to reach for something else
- Open weights are promised but were not published at launch
- No full model card or reproducible benchmark configuration yet
- As a Chinese model it answers cautiously on politically sensitive topics
- Response latency grows noticeably at the far end of the context window
- No image or video generation — it only accepts them as input
How Qwen3.8 Max responds
Qwen3.8 Max
Four scenarios where it pays for itself
More about Qwen3.8 Max
Qwen3.8 Max — Alibaba's Flagship AI Model
Qwen3.8 Max is the most capable model in the Qwen family as of August 2026. It packs 2.4 trillion parameters into a sparse Mixture-of-Experts architecture, activating roughly 95 billion per request. That design keeps latency reasonable for a model of this size while landing well below Western flagships on price.
What Qwen3.8 Max can do
The model is natively multimodal — it accepts text, images and video as input. The context window is 1,000,000 tokens, enough for an entire mid-sized codebase or around 750 pages of text. Maximum output runs to 131,072 tokens, so it can produce a substantial module in a single pass.
Alibaba built this release around agentic work. In internal testing, Qwen3.8 Max sustained autonomous software development for more than ten days straight — writing code, running its own tests, reading logs and iterating without human input. It scores 86.6 on Terminal-Bench 2.1 and 67.7 on SWE-bench Pro.
Multilingual strength
Qwen has always been strong across languages, and this release continues that. It handles documents, code comments and long conversations in dozens of languages without the translated-from-English feel that weaker models produce.
Try it on NetRoom
On NetRoom you can run Qwen3.8 Max straight from the browser — no Alibaba Cloud account, no separate API key, no setup. Billing is per token actually used, so you pay for output rather than a subscription tier.
Who it's for
Developers who need long context and agentic workflows; analysts working through large document sets; teams that care about cost per million tokens rather than the logo on the box.
Version history of Qwen3.8 Max
| Version | Date | What changed |
|---|---|---|
| 3.8 Max current |
|
|
| 3.6 Max Preview |
|
Try Qwen3.8 Max
right now
Free access to basic models. No card, no obligations.