Qwen 3.5

Qwen 3.5 — explained in more detail

Qwen 3.5 is Alibaba’s model family rolled out in February/March 2026. The architecture combines dense models at smaller sizes with Mixture-of-Experts from 27B upward. Multimodality is integrated directly into the architecture via a “Unified Vision-Language Foundation” with early-fusion training. All variants are open-weights.

Variants and release dates

• Qwen 3.5 Flagship & Plus (February 16, 2026): top tier including the 397B-A17B MoE — 397B total, 17B active per token. 8–19× faster decoding at roughly 60% lower cost than Qwen3-Max.
• Medium series (February 24, 2026): Qwen3.5-122B-A10B, Qwen3.5-35B-A3B, Qwen3.5-27B. The 35B-A3B flagship of this tier exceeds 1M tokens of context.
• Small series (March 2, 2026): Qwen3.5-0.8B / 2B / 4B / 9B — built for on-device applications. 262K native context, extended to 1M on the 9B model.

Benchmarks and context

The 9B model scores 82.5 on MMLU-Pro, 81.7 on GPQA Diamond, 55.2 on LongBench v2 — beating GPT-OSS-120B as well as its own 3× larger predecessor Qwen3-30B. Alibaba positions the open-weights line as direct competition to Claude Sonnet 4.5 and GPT-5 Mini, with much cheaper self-hosting as the differentiator.

More on this topic

See everything in one place:Qwen→

Encyclopedia · 16 May 2026Comparing AI models — who builds what and how to chooseThe major model families in 2026 at a glance. Who builds Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen — and which model to pick when.Encyclopedia · 16 May 2026LLM Model Families at a GlanceClaude, GPT, Gemini, Llama & co. — who builds what, where each family shines, and how to pick the right model for your own use case.News · 29 Aug 2026Open-weight wave July/August 2026: GLM-5.3, DeepSeek V4-Pro, Qwen3.8July to August 2026 brought a wave of open models: GLM-5.3, DeepSeek V4-Pro, Qwen3.8. What the permissive licenses mean for self-hosting.