MetaMeta　·　当前代current generation Llama 4Llama 4（2025-04）

LlamaLlama模型家族谱系卡model family lineage card

全球生态最大的开放权重家族：微调、量化、工具链覆盖最全，但许可是带归属条款（Built with Llama）与规模条款（7 亿 MAU）的社区许可，商用前读全文。The largest open-weight ecosystem by far — fine-tunes, quantisations and tooling are everywhere — but the licence is a community licence with attribution ("Built with Llama") and scale (700M MAU) clauses. Read it fully before commercial use.

本站未收录 Llama 系列的量化对比：Llama-4-Scout 的第三方 GGUF 下载量约 12.8 万（2026-10-30 复测），未过 METHODOLOGY 的量化生态门槛（约 30 万）。要看量化对比，去 qwen3-8 / gemma-4 系列。This site does not cover Llama quantisations: Llama-4-Scout's third-party GGUF downloads (~128k, re-measured 2026-10-30) fall below the METHODOLOGY ecosystem threshold (~300k). For quantisation coverage, see the qwen3-8 / gemma-4 series.

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　国外出品foreign-made　出品方为 Meta。Made by Meta.
国内下载Direct access　直连未核验direct unknown　ModelScope 站内检索 Llama 未命中官方镜像（2026-10-20）；Hugging Face 直连不可达——两条主路都不可达，故记 unknown 而不是 no（可能有本站未检索到的渠道）。No official Llama mirror found on ModelScope (searched 2026-10-20); Hugging Face direct unreachable — both main routes unreachable, so unknown rather than no (a route we have not found may exist).
备案状态CAC filing　未见于备案名单not in filing list：公示名单未检索到 Meta / Llama 相关条目（2026-10-20）。No Meta / Llama entry found in the public registry (searched 2026-10-20).

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.

网信办《生成式人工智能服务已备案信息》滚动公告页（13 个批次附件 PDF，2026-10-20 全量下载检索）：CAC · filed generative-AI service information (rolling announcement page, 13 batch attachment PDFs, full download & search on 2026-10-20): https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm

代次Generations

Llama 4Llama 42025-04当前代current

当前代。两个 MoE 规格：Scout（17B-16E）与 Maverick（17B-128E），均为图像-文本多模态输入。Current generation. Two MoE builds: Scout (17B-16E) and Maverick (17B-128E), both with image-and-text multimodal input.
官方页面 ↗official page ↗
Llama 3.3Llama 3.32024-11

上一代密集旗舰 70B-Instruct。Previous dense flagship, 70B-Instruct.
官方页面 ↗official page ↗
Llama 3.2 / 3.1Llama 3.2 / 3.12024-09

小杯视觉线（1B / 3B）与 405B 大杯线（3.1）；当前生态里大量微调的源头。The small vision line (1B / 3B) and the 405B flagship (3.1); the origin of much of today's fine-tune ecosystem.
官方页面 ↗official page ↗

尺寸档Size tiers

小杯 1B / 3B：端侧与视觉小杯Small 1B / 3B: on-device and vision-small
中杯 8B / 70B：本地部署常用档Medium 8B / 70B: the common local-deployment tiers
大杯 405B / 17B-128E MoE：服务器级Large 405B / 17B-128E MoE: server-class

许可Licences

otherLlama 4 Community License（license_name: llama4）——需登录接受条款；HF 页可逐条读许可全文Llama 4 Community License (license_name: llama4) — gated, terms accepted on sign-in; full text readable on the HF page 许可原文 ↗licence text ↗

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

HF 模型卡 metadata：Llama-4 两个规格均为 image-text-to-text 管线（图像+文本输入），支持 12 种语言（ar/de/en/es/fr/hi/id/it/pt/th/tl/vi）。HF model-card metadata: both Llama-4 builds are image-text-to-text (image + text input) and list 12 languages (ar/de/en/es/fr/hi/id/it/pt/th/tl/vi).出处 ↗source ↗
许可原文（HF 门控页）：分发衍生作品须「prominently display "Built with Llama"」；月活超过 7 亿的产品须另行向 Meta 申请许可（Additional Commercial Terms 节）。Licence text (gated HF page): derivatives must "prominently display 'Built with Llama'"; products above 700M MAU need a separate licence from Meta (Additional Commercial Terms).出处 ↗source ↗
官方未说明（本次核验）：Llama-4 的上下文长度未在可直抓的官方页给出——模型卡 README 为门控内容，llama.com 页有身份确认拦截。本站不转述未经核验的数字。Not officially stated (this check): Llama-4 context length was not available on publicly fetchable official pages — the model-card README is gated and llama.com sits behind an identity check. No unverified numbers are repeated here.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

Llama-4-Scout-17B-16E-Instruct 模型页（含门控许可全文，2026-10-20 直抓 200）：Llama-4-Scout-17B-16E-Instruct model page (includes full gated licence, direct fetch 200 on 2026-10-20): https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct
Llama-4-Maverick-17B-128E-Instruct 模型页（2026-10-20 直抓 200）：Llama-4-Maverick-17B-128E-Instruct model page (direct fetch 200 on 2026-10-20): https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct
llama.com/llama4 官方页（2026-10-20 可达但对自动抓取有身份确认，未能取正文）：llama.com/llama4 official page (reachable 2026-10-20 but identity-checks automated fetches; body not retrievable): https://www.llama.com/llama4/
Llama-3.3-70B-Instruct 模型页（代次核验，2026-10-20 直抓 200）：Llama-3.3-70B-Instruct model page (generation check, direct fetch 200 on 2026-10-20): https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct
Llama-3.1-405B 模型页（代次核验，2026-10-20 直抓 200）：Llama-3.1-405B model page (generation check, direct fetch 200 on 2026-10-20): https://huggingface.co/meta-llama/Llama-3.1-405B

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.