· Updated
Qwen3.5-35B-A3B, Qwen3.5-122B-A10B, and Qwen3.5-27B. The headline result is the 35B-A3B: a sparse mixture-of-experts model that activates just 3 billion parameters per forward pass, yet matches or exceeds the previous 235B flagship across SWE-bench Verified, GPQA Diamond, and agentic tool-use benchmarks.The architecture combines Gated Delta Networks for linear attention with sparse MoE layers, delivering high throughput at a fraction of the inference cost. All models support native vision-language understanding through early fusion training and 201 languages out of the box. Qwen3.5-Flash, the hosted production version, extends context to 1 million tokens.
Weights for the open models are available on Hugging Face, compatible with vLLM, SGLang, and Transformers.
🚀 Introducing the Qwen 3.5 Medium Model Series Qwen3.5-Flash · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-27B ✨ More intelligence, less compute. • Qwen3.5-35B-A3B now surpasses Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B — a reminder that better architecture, data quality, https://t.co/ZWPibMn6at
More like this
QwenApr 30
Fireworks AIApr 30
Arena17h ago
Arena.aiApr 21