Alibaba Releases Qwen3.8-27B: Open 27B Model Comparable to Frontier...

Alibaba Releases Qwen3.8-27B: Open 27B Model Comparable to Frontier in Agentic Coding

On August 14, 2026, Alibaba released Qwen3.8-27B — a 27.8 billion parameter dense model under Apache-2.0 that understands text, images, and video, and is comparable to Opus4.6 Max in agentic coding according to official benchmarks.

Compiled by Sergey KostenchukPublished 2026-08-14Updated 2026-09-18

2026-08-14 Research HuggingFace

🤖 Alibaba Releases Qwen3.8-27B: Open 27B Model Comparable to Frontier in Agentic Coding

On August 14, Qwen (Alibaba) released Qwen3.8-27B — a 27.8 billion parameter dense model under Apache-2.0 that understands text, images, and video. According to official benchmarks, it outperforms the closed Opus4.6 Max in SWE-bench Pro (61.7 vs 53.4) and OSWorld-Verified (84.3 vs 72.7), but lags in HLE (30.8 vs 40.0).

🌍 An open 27B model that can run on a single server reaches the level of closed frontier models in agentic coding and computer use: self-hosted becomes a real alternative to frontier APIs, and price pressure on paid APIs grows.

👤 Weights are available on Hugging Face (Apache-2.0, ~55.6 GB in BF16). Running on vLLM/SGLang: native context 262K, controllable 'thoughtfulness' via reasoning_effort. AMD provided day-0 support on Ryzen AI Max and Radeon on release day.

Source 1: https://huggingface.co/Qwen/Qwen3.8-27B

Sources