Qwen3.5-27B
Qwen3.5-27B is an AI model published by 阿里巴巴, released on 2026-02-25, for 推理大模型, with 270.0B parameters, and 1010K tokens context length, requiring about 55.6 GB storage, under the Qwen License license.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Qwen3.5-27B currently shows benchmark results led by Pinch Bench (2 / 37, score 90), IF Bench (2 / 27, score 76.50), MMLU Pro (11 / 116, score 86.10). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.
Qwen3.5-27B是阿里云于2026年2月25日发布的一款密集架构(Dense)开源模型,旨在为用户提供一种轻量化、高能效的大模型选择[citation:1][citation:3]。该模型拥有270亿参数,专注于在保持高性能的同时,降低模型部署和使用的门槛。
尽管尺寸相对较小,Qwen3.5-27B在多项任务中表现出色,尤其在交互、编程、长文本理解和数学推理方面展现了强大能力[citation:2]。在SWE-bench Verified编程能力测评中,该模型取得了最高分,证明了其在真实软件工程问题上的解决能力[citation:2]。此外,它在多模态幻觉抑制与事实一致性方面也有优异表现[citation:2]。
开发者可以通过Hugging Face、GitHub或魔搭社区下载该模型进行本地部署和微调[citation:8]。这款模型特别适合资源受限的场景,或作为需要快速响应和高性价比的特定任务的专用模型。
欢迎关注 DataLearner 官方微信,获得最新 AI 技术推送
