Alibaba's Qwen3.8-27B lands under Apache 2.0 at 73.0 on Terminal-Bench 2.1

Tongyi Lab released Qwen3.8-27B under Apache 2.0 — a 27.8B-parameter vision-language model with a 262,144-token native context that scores 73.0 on Terminal-Bench 2.1, up from 63.4 for Qwen3.6-27B. At 14-17GB in 4-bit it fits a single 24GB GPU, putting agentic coding at that benchmark tier on one desktop card rather than a hosted frontier API.

Source: huggingface.co ↗

Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability.

Tongyi Lab

Why this matters

→ 27B model fits on single 24GB GPU; desktop access to agentic coding benchmark tier
→ Apache 2.0 license enables commercial deployment without vendor API dependency
→ 262K native context handles complex multi-step tasks end-to-end

Desktop-class agentic AI