OPEN SOURCE OBSERVER
从项目热度走向工程判断：追踪开源工具、模型与开发者生态。
01 / CODE
| 排名 | 项目 | 语言 | 今日新增 | 总 Star |
|---|---|---|---|---|
| 01 | lnkiai/m3e-canvasSketch Material 3 Expressive screens in the browser and turn them into vibe-coding prompts. | TypeScript | — | 7,303 |
| 02 | MengTo/threeuiOpen-source ThreeUI Community catalog with live interactive components and complete Community source. | HTML | — | 5,906 |
| 03 | sapientinc/PRAXISTAutonomous research system for measurable, computer-executable research. | Python | — | 5,717 |
| 04 | rakanki911/DLSS5-SwapperInstall, tune and restore DLSS 5 in your games with one click. Native RenoDX, DLSS5-Feeder for games without DLSS, OptiScaler and multipass routes; DirectX 8/9/11/12, Vulkan, OpenGL, DirectDraw and emulators. In-game F8 overlay, automatic backups, and a community page showing what works on your games and your graphics card. | JavaScript | — | 5,543 |
| 05 | XiaoDuoYa/codex-with-chatgptChatGPT thinks. Codex works. Use ChatGPT as the planning brain while keeping the Codex harness. | TypeScript | — | 5,185 |
| 06 | crmne/spotifastSpotify, native and fast. One lightweight Rust app for your whole library, local playback, and Spotify Connect on Linux, macOS, and Windows. | Rust | — | 4,330 |
| 07 | HEJustinSun/my-girlfriend-jingtian-latex暂无简介 | TeX | — | 4,210 |
| 08 | bojieli/ai-infra-book《深入理解 AI Infra：量化分析与系统设计》（李博杰 著）开源书稿：从硬件约束和模型架构出发，量化推导 LLM 推理与训练系统设计。含全书正文、PDF、配套计算工具与实验 | Python | — | 4,123 |
| 09 | ashemag/human-atlasOpen-source 3D anatomy explorer: 2,234 selectable BodyParts3D meshes, system layers, search, and exploded views. | TypeScript | — | 3,603 |
| 10 | b-nnett/grok-bot-0.18-reconstructedUnofficial source-oriented reconstruction and extension of Grok Bot 0.18.0 for macOS | TypeScript | — | 3,506 |
02 / MODELS
| 排名 | 模型 | 任务 | 收藏 | 下载 |
|---|---|---|---|---|
| 01 | Z-Image-Turbo | 文本生成图片 | 2,524 | 811,957 |
| 02 | MiniMax-H3 | — | 2,002 | 145,860 |
| 03 | Qwen3.8-27B | 视觉多模态理解 | 1,473 | 388,453 |
| 04 | DeepSeek-R1 | — | 1,384 | 927,369 |
| 05 | Qwen3.8-Flash-Next | 视觉多模态理解 | 1,217 | 58,474 |
| 06 | Qwen3-Coder-480B-A35B-Instruct | 文本生成 | 1,095 | 138,499 |
| 07 | IndexTTS-2 | 语音合成 | 981 | 208,617 |
| 08 | chatglm3-6b | — | 830 | 1,590,840 |
| 09 | Qwen3.8-2.4T-A95B | 文本生成 | 770 | 1,955 |
| 10 | Qwen-7B-Chat | 文本生成 | 757 | 693,305 |
DEVELOPER VIEW
对开发者最直接的价值有三点。一是免费且可自托管的选择变多了：Qwen 系列从 7B 到 2.4T 有完整梯度，chatglm3-6b、DeepSeek-R1 仍在榜，本地部署和微调的成本门槛持续下降。二是新范式可抄：codex-with-chatgpt 的"规划脑 + 执行器"分离、m3e-canvas 的"草图转 prompt"，都是可以直接搬进自己产品的工作流设计。三是系统工程知识可量化：ai-infra-book 附带计算工具，适合想搞清楚显存、吞吐、并行策略边界的工程师。
WORTH A STAR
bojieli/ai-infra-book：少见的把系统设计做成可推导、可实验的开源书稿，配合 PDF 与计算工具，对做推理优化和训练调度的工程师有长期价值。
sapientinc/PRAXIST：如果自主研究系统要真正落地，可测量、可执行是绕不开的门槛，这个方向值得持续跟踪。
lnkiai/m3e-canvas：把 UI 草图直接转成 vibe-coding 提示，思路简单但切中"意图到实现"的关键一环，适合作为 AI 设计工具的参考实现。
SOURCES & METHOD
GitHub 搜索 API（近 30 天新建 · 按 Star 排序）（项目热度、语言、Star） · ModelScope（模型任务、收藏与下载）
抓取日期：2026-09-18。排名是当日快照，指标可能随平台更新而变化。