Qwen 3.6 / 3.7-Max / 3.8 (2026 line)
Alibaba's 2026 wave, topping out at a 2.4-trillion-parameter open MoE.
You're looking at an AI model record: who built it, what it can do, and the hardware and data behind it.
Alibaba kept the world's most-forked open family sprinting through 2026: Qwen 3.6 (April — a 35B-A3B mixture-of-experts and a 27B dense model, Apache 2.0), the API-only closed Qwen 3.7-Max (May 19), and in August the Qwen 3.8 generation — headlined by a 2.4T-parameter MoE activating 95B with a hybrid DeltaNet attention design, alongside an open 27B dense vision-language model. The pattern mirrors the industry's split: permissive open weights for the family, API-only for the very top.
Specifications & key facts
Each figure is labeled by basis: vendor spec, vendor claim, reported, estimate, or official. "Not publicly disclosed" means exactly that — we never invent values.
Overview
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Developed by
Successor to
Competes with
Sources
Open-source LLM landscape 2026 (Qwen line) ↗ · Chinese AI models 2026 comparison ↗
Record reviewed as of 2026-08-31 — see Methodology.