跳到主要内容Skip to main content
投机取巧 Speculative Speculation

阿里巴巴 · 通义千问Alibaba · Qwen　·　当前代current generation Qwen3.8Qwen3.8（2026-08）

千问 QwenQwen模型家族谱系卡model family lineage card

中文与代码口碑最强的开放权重家族之一；Apache-2.0 主力规格许可干净。本站量化对比的基准系列（qwen3-8）就在这一代。One of the strongest open-weight families for Chinese and code, with clean Apache-2.0 licensing on mainline builds. This site's benchmark quantisation series (qwen3-8) sits in this generation.

本站已收录该家族的量化对比：This site covers this family's quantisations: 千问 3.8Qwen3.8

先说清楚：这个站的数据怎么来的First, where this site's data comes from

✅ 实测字段Measured fields

直接从 Hugging Face 公开 API 读取的客观数据：文件体积（字节）、上下文长度、架构、许可、下载量、更新时间。Objective data read straight from the Hugging Face public API: file sizes in bytes, context length, architecture, licence, downloads, last updated.

✅ 推导Derived

由实测字段按公开规则算出：每权重比特数（依 llama.cpp 命名标准）、档位归属、显存需求区间。推导规则在Computed from the measured fields by published rules: bits per weight (per the llama.cpp naming convention), tier assignment, VRAM requirement range. The rules are fully published in METHODOLOGY.md 完整公开，任何人可复算。 — anyone can recompute them.

❌ 我们不说的What we do not claim

哪个量化质量更高、哪个更快、哪个模型更强——这些需要实测，本站不跑 benchmark，所以不装作知道。凡我们给出的主观判断，一律显式标注「我们的口径」。Which quantisation is higher quality, which is faster, which model is stronger — these need measurement. This site does not run benchmarks, so it does not pretend to know. Every subjective judgement we do offer is explicitly marked as our own reading.

本站This site does 不下载、不托管任何模型权重文件not download or host any model weight files。所有数字都来自公开元数据，点击即可回到发布页核对。. Every figure comes from public metadata, and clicking it takes you back to the release page to verify. 采集时间Collected: 2026-10-30（数据快照，非实时） (a snapshot, not live)

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-20

  • 出品属地Origin　国产domestic　出品方为阿里巴巴 · 通义千问（杭州）。Made by Alibaba Qwen (Hangzhou).
  • 国内下载Direct access　国内可直连direct OK　ModelScope 有 Qwen 官方组织镜像（2026-10-20 实测 200）；Hugging Face 直连不可达。Official Qwen org mirror on ModelScope (measured 200 on 2026-10-20); Hugging Face direct unreachable.　·　国内镜像入口domestic mirror
  • 备案状态CAC filing　已备案Filed：ZheJiang-TongYiQianWen-20230901　公示名称listed as「通义千问大模型」　·　备案时间filed 2023-09-12　·　首批第 11 行。Row 11 of the first batch.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「国内可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise.

代次Generations

  • Qwen3.8Qwen3.82026-08当前代current

    当前代。27B 密集 + 2.4T-A95B MoE + Flash-Next（180B MoE）三线；27B 密集版是本地部署圈热度最高的一档。Current generation. Three lines: 27B dense, 2.4T-A95B MoE, and Flash-Next (180B MoE). The 27B dense build is the most talked-about for local deployment.

    官方页面 ↗official page ↗
  • Qwen3.5Qwen3.52026

    上一代的架构基础——官方 Qwen3.8 README 原文：「Built on the architectural foundation of Qwen3.5」。Architectural foundation of the previous generation — the official Qwen3.8 README states: "Built on the architectural foundation of Qwen3.5".

    官方页面 ↗official page ↗
  • Qwen3Qwen32025-04

    更早一代的主力，0.6B 到 235B-A22B 全尺寸覆盖，apache-2.0 与混合思考模式从这一代立住。Earlier flagship generation, spanning 0.6B to 235B-A22B; Apache-2.0 licensing and hybrid thinking modes were established here.

    官方页面 ↗official page ↗

尺寸档Size tiers

  • 小杯 0.6B–9B：端侧、免费额度、 CPU 可跑Small 0.6B-9B: on-device, free tiers, CPU-capable
  • 中杯 14B–35B：本地部署主力（27B 密集为当前代代表）Medium 14B-35B: the local-deployment workhorse (27B dense leads this generation)
  • 大杯 180B–2.4T：MoE 巨杯，需服务器级显存Large 180B-2.4T: MoE flagships needing server-class VRAM

许可Licences

官方能力声明（带出处）Official capability claims (with sources)

  • 官方 README：Qwen3.8-27B 是原生视觉-语言模型（「a native vision-language model that understands images and videos」），支持从 STEM 图表到小时级视频的理解。Official README: Qwen3.8-27B is "a native vision-language model that understands images and videos", covering STEM diagrams up to hour-long video.出处 ↗source ↗
  • 官方 README：思考模式默认开启、可按请求关闭，推理深度用 reasoning_effort 调节（主语是托管版；权重版字段以本页配置为准）。Official README: thinking mode is on by default and can be disabled per request, with reasoning depth tuned via reasoning_effort (stated for the hosted version).出处 ↗source ↗
  • 官方 README 提到的托管版「1M context length by default」是云端服务声明，不直接等于权重版上限；权重版上下文以 config 字段为准。The README's "1M context length by default" is a statement about the hosted service, not automatically the weights build; check config for the weights.出处 ↗source ↗

以上均为厂商官方页面原文或官方自评基准的转述；涉及横评结论的，本站不背书——本站不跑 benchmark。The above are quotations of official pages or transcriptions of vendor-run benchmarks; this site does not endorse cross-model conclusions — it runs no benchmarks.

出处Sources

核验快照：2026-10-30 · 本站不实测、不排名，未知就写未知。Verification snapshot: 2026-10-30 · no benchmarks, no rankings — unknown stays unknown.