Qwen 3.5 small models bring open AI to consumer GPUs

Alibaba has expanded its Qwen 3.5 lineup with a series of smaller models aimed at running on everyday hardware. The family now includes 2B, 7B, 14B, 27B, and 35B variants, along with quantized builds that compete with GPT “Mini” and Claude Sonnet on standard benchmarks.

A 27B FP8 version weighs in at around 31 GB, while community-made 4-bit quantized builds drop VRAM requirements to roughly the 10 GB range. That makes it possible to run what the community considers “frontier-level” open models on a single high-end consumer GPU.

For developers and enthusiasts, Qwen 3.5’s smaller variants offer a way to experiment with strong language and multimodal capabilities without relying on cloud-hosted APIs. Companies like AMD have already provided articles concerning Day 0 support for Qwen 3.5.