Qwen3.8 27B

Alibaba's Qwen team released Qwen3.8-27B on August 14, 2026 as an open-weight Apache 2.0 multimodal model. The checkpoint contains 27.78 billion parameters, accepts text, images and video, and ships with a native 262,144-token context window, positioning it as a strong dense candidate for local deployment near 30B parameters. Qwen reports substantial gains over Qwen3.6-27B in agentic coding, computer use, and vision-language tasks, citing Terminal-Bench 2.1 rising from 63.4 to 73.0, DeepSWE 1.1 from 13.3 to 42.2, OSWorld-Verified from 63.9 to 84.3, and SWE-MM from 25.7 to 38.6. Kingy.ai notes these numbers are Qwen-sourced, several benchmarks are in-house or modified, and longer contexts beyond 262K require YaRN scaling rather than the native window.

Artificial Analysis profiles Qwen3.8 27B (xhigh reasoning variant) as an August 2026 open-weights release from Alibaba, confirming the 27B-parameter size, 256K context window, text/image/video inputs, and Apache 2.0 license. The model scores 34 on the Intelligence Index, ranking first among 142 small open-weights models in its class. Independent evaluation reports 44 output tokens per second (notably slow for its class) and pricing of $0.50 per million input and $3.00 per million output tokens with an 80% cache discount. Qwen3.8 27B is described as particularly expensive relative to similar open-weight models and very verbose, generating 200 million tokens during the Intelligence Index run versus a class median of 82 million.