{ "categories": [ "工具箱·终端", "工具箱·国产", "本地部署", "知识库与索引", "站点与自研", "厂商·国产", "厂商·美国", "厂商·欧洲", "编程套餐", "对比合集" ], "columns": [ "工具箱·终端", "工具箱·国产", "本地部署", "知识库与索引", "站点与自研", "厂商·国产", "厂商·美国", "厂商·欧洲", "编程套餐", "对比合集" ], "records": [ { "id": "X-001", "title": "Claude Code", "en": "CLAUDE CODE", "department": "终端工具", "category": "工具箱·终端", "date": "已上线", "lead": "Dario Amodei（Anthropic 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "Anthropic 官方终端编程代理，npm 国内镜像可直连安装。", "findings": [ "国内两条安装路线：npm 镜像优先，官方安装器需网络环境支持。" ], "source": "/docs/claude-code", "link": "#/docs/claude-code", "slot": null }, { "id": "X-002", "title": "Codex CLI", "en": "CODEX CLI", "department": "终端工具", "category": "工具箱·终端", "date": "已上线", "lead": "Sam Altman（OpenAI CEO）、Greg Brockman（总裁）", "clearance": "REFERENCE AREA", "abstract": "OpenAI 官方终端编程代理，一行命令安装，与微软商店无关。", "findings": [ "国内两条安装路线：npm 镜像优先；官方 chatgpt.com 脚本在受限网络下不可直连。" ], "source": "/docs/codex", "link": "#/docs/codex", "slot": null }, { "id": "X-003", "title": "Google Antigravity", "en": "ANTIGRAVITY", "department": "IDE", "category": "工具箱·终端", "date": "已上线", "lead": "Demis Hassabis（联合创始人兼 CEO）、Koray Kavukcuoglu（SVP）", "clearance": "REFERENCE AREA", "abstract": "谷歌 AI 编程 IDE，官网直下安装包；Gemini CLI 免费档已并入其内置 CLI。", "findings": [ "Gemini CLI 免费档用户直接装 Antigravity，它已内置 Antigravity CLI。" ], "source": "/docs/antigravity", "link": "#/docs/antigravity", "slot": null }, { "id": "X-004", "title": "CC Switch", "en": "CC SWITCH", "department": "桌面端", "category": "工具箱·终端", "date": "已上线", "lead": "Jason Young（farion1231，项目作者）", "clearance": "REFERENCE AREA", "abstract": "六大编程客户端的供应商切换台，key 全程留在本机。", "findings": [ "切换是全量重写配置文件，别在配置里手工放自定义字段。" ], "source": "/docs/cc-switch", "link": "#/docs/cc-switch", "slot": null }, { "id": "X-005", "title": "OpenCode", "en": "OPENCODE", "department": "CLI · 开源", "category": "工具箱·终端", "date": "已上线", "lead": "Jay V、Frank Wang（SST 团队，项目作者）", "clearance": "REFERENCE AREA", "abstract": "开源终端代理，支持任意 LLM 供应商与本地模型。", "findings": [ "beta 通道升级可能清空数据，重要会话别放 beta。" ], "source": "/docs/opencode", "link": "#/docs/opencode", "slot": null }, { "id": "X-006", "title": "Gemini CLI", "en": "GEMINI CLI", "department": "CLI · 迁移", "category": "工具箱·终端", "date": "已上线", "lead": "Demis Hassabis（DeepMind 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "谷歌终端代理；免费档自 2026-06 起迁移至 Antigravity CLI。", "findings": [ "免费账号请直接装 Antigravity，别按老教程硬装 Gemini CLI。" ], "source": "/docs/gemini-cli", "link": "#/docs/gemini-cli", "slot": null }, { "id": "X-007", "title": "Aider", "en": "AIDER", "department": "CLI · 开源", "category": "工具箱·终端", "date": "已上线", "lead": "Paul Gauthier（项目作者）", "clearance": "REFERENCE AREA", "abstract": "老牌终端结对编程，Git 深度集成、自动提交。", "findings": [ "项目已进入维护模式（2025-08 后无功能更新），稳定但新特性少。" ], "source": "/docs/aider", "link": "#/docs/aider", "slot": null }, { "id": "X-008", "title": "Cline", "en": "CLINE", "department": "插件 · 开源", "category": "工具箱·终端", "date": "已上线", "lead": "saoudrizwan（项目作者）", "clearance": "REFERENCE AREA", "abstract": "VS Code / CLI 自主编程代理，Plan/Act 双阶段工作流。", "findings": [ "热门 fork Roo Code 已归档停更，别按旧教程去装。" ], "source": "/docs/cline", "link": "#/docs/cline", "slot": null }, { "id": "X-009", "title": "Cursor", "en": "CURSOR", "department": "IDE · AI 原生", "category": "工具箱·终端", "date": "已上线", "lead": "Michael Truell（Anysphere 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "AI 原生编辑器代表：补全 / 改写 / Agent 模式并存，VS Code 用户迁移成本最低。", "findings": [ "编辑器体验是它的核心增量：补全与改写在日常编码里比代理模式更高频。", "订阅内额度用完后按量或降级，重度使用会「月初爽、月中卡」，先看官网当前定价页。", "依赖其云端服务：网络不稳时体验明显下降，接入本地模型能力有限。", "与 Windsurf 的取舍见站内对比《Cursor vs Windsurf》；预算敏感可换 VS Code 插件 + 本地模型。" ], "source": "/docs/cursor", "link": "#/docs/cursor", "slot": null }, { "id": "X-010", "title": "Qwen Code", "en": "QWEN CODE", "department": "CLI · 阿里", "category": "工具箱·国产", "date": "已上线", "lead": "吴泳铭（阿里集团 CEO）、周靖人（阿里云 CTO，代管通义）", "clearance": "REFERENCE AREA", "abstract": "阿里 Qwen 官方 CLI，基于 Gemini CLI 框架重构。", "findings": [ "npm 包 @qwen-code/qwen-code 走镜像源直装；OAuth 登录可领免费额度。" ], "source": "/docs/qwen-code", "link": "#/docs/qwen-code", "slot": null }, { "id": "X-011", "title": "GLM Coding", "en": "GLM CODING", "department": "套餐 · 智谱", "category": "工具箱·国产", "date": "已上线", "lead": "张鹏（智谱 CEO）、唐杰（创始人）", "clearance": "REFERENCE AREA", "abstract": "智谱编程套餐，兼容 Claude Code 接入方式。", "findings": [ "Anthropic 兼容接入 Claude Code / CC Switch；base url 必须用官方专用路径，随手改会 404。" ], "source": "/docs/glm-coding", "link": "#/docs/glm-coding", "slot": null }, { "id": "X-012", "title": "Kimi Code CLI", "en": "KIMI CODE", "department": "CLI · Kimi", "category": "工具箱·国产", "date": "已上线", "lead": "杨植麟（月之暗面创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "月之暗面官方 CLI（单文件版，无需 Node.js），K 系模型第一梯队。", "findings": [ "npm 上的 kimi-cli 是无关旧抢注包，认准 code.kimi.com 官方脚本。" ], "source": "/docs/kimi-cli", "link": "#/docs/kimi-cli", "slot": null }, { "id": "X-013", "title": "DeepSeek 接入", "en": "DEEPSEEK", "department": "API", "category": "工具箱·国产", "date": "已上线", "lead": "梁文锋（DeepSeek 创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "无官方 CLI，经 Anthropic 兼容端点接入各编程客户端。", "findings": [ "Anthropic 兼容端点须按官方文档的专用路径填写，不是直接把 base url 改成主域。" ], "source": "/docs/deepseek", "link": "#/docs/deepseek", "slot": null }, { "id": "X-014", "title": "CodeBuddy Code", "en": "CODEBUDDY", "department": "腾讯", "category": "工具箱·国产", "date": "已上线", "lead": "马化腾（腾讯创始人）、姚顺雨（首席 AI 科学家）", "clearance": "REFERENCE AREA", "abstract": "腾讯云编程三形态：IDE / 插件 / CLI，混元模型驱动，国内直连。", "findings": [ "CLI 的 Web UI 模式会占本地端口，内网环境需要放行；国内版与国际版账号额度不通用。" ], "source": "/docs/codebuddy", "link": "#/docs/codebuddy", "slot": null }, { "id": "X-015", "title": "MiniMax Code CLI", "en": "MINIMAX CODE", "department": "CLI · MiniMax", "category": "工具箱·国产", "date": "已上线", "lead": "闫俊杰（MiniMax 创始人、董事长兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "MiniMax 终端代理 mcode，可一键配置进其他编程客户端。", "findings": [ "国内版与国际版账号、key、额度分区独立，别混用。" ], "source": "/docs/minimax", "link": "#/docs/minimax", "slot": null }, { "id": "X-016", "title": "Trae", "en": "TRAE", "department": "IDE · 字节", "category": "工具箱·国产", "date": "已上线", "lead": "梁汝波（字节跳动 CEO）、吴永辉（Seed 基础研究负责人）", "clearance": "REFERENCE AREA", "abstract": "字节的 AI 原生 IDE（个人版有免费额度），主打「边说边改」的对话式编码。", "findings": [ "资源占用偏高，老机器上比 VS Code 系更吃内存，先试个人版再决定。", "历史上的遥测争议：介意数据边界的话优先用它对应的开源代理形态（站内 Trae Agent 档）。", "国内直连可用性比海外 IDE 好，是它相对 Cursor / Windsurf 的实际优势。" ], "source": "/docs/trae", "link": "#/docs/trae", "slot": null }, { "id": "X-017", "title": "百度 Comate", "en": "COMATE", "department": "插件 · 百度", "category": "工具箱·国产", "date": "已上线", "lead": "李彦宏（百度创始人、董事长兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "文心快码：插件 / IDE / CLI 三形态，个人版有免费额度。", "findings": [ "CLI 包名 @comate/comatecli 已实测；免费版额度按月刷新，重度使用会触顶降级。" ], "source": "/docs/comate", "link": "#/docs/comate", "slot": null }, { "id": "X-018", "title": "Qoder", "en": "QODER", "department": "IDE · 阿里", "category": "工具箱·国产", "date": "已上线", "lead": "吴泳铭（阿里集团 CEO）", "clearance": "REFERENCE AREA", "abstract": "阿里现役编码产品：由通义灵码整合而来（2026），Credits 额度制、自主度更高。", "findings": [ "由通义灵码品牌整合而来：老用户按官方迁移说明切到 Qoder。", "Credits 计费下「高自主度任务」消耗快，续费前看清换算口径。", "与阿里云账号体系打通，企业侧可走百炼；个人先试免费额度再决定。" ], "source": "/docs/qoder", "link": "#/docs/qoder", "slot": null }, { "id": "X-019", "title": "智谱 CodeGeeX", "en": "CODEGEEX", "department": "插件 · 智谱", "category": "工具箱·国产", "date": "已上线", "lead": "张鹏（智谱 CEO）、唐杰（创始人）", "clearance": "REFERENCE AREA", "abstract": "多语言代码助手；产品重心已迁移 ZCode。", "findings": [ "插件可能随时停更，别在它上面搭长期工作流。" ], "source": "/docs/codegeex", "link": "#/docs/codegeex", "slot": null }, { "id": "X-020", "title": "讯飞 iFlyCode", "en": "IFLYCODE", "department": "插件 · 讯飞", "category": "工具箱·国产", "date": "已上线", "lead": "刘庆峰（科大讯飞创始人、董事长兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "星火飞码，个人额度偏紧，强项在政企私有化场景。", "findings": [ "强项在政企私有化部署；个人免费额度偏紧，个人开发者可优先 Qwen Code / GLM Coding。" ], "source": "/docs/iflycode", "link": "#/docs/iflycode", "slot": null }, { "id": "X-021", "title": "ComfyUI", "en": "COMFYUI", "department": "画图 · 视频", "category": "本地部署", "date": "已上线", "lead": "comfyanonymous（原作者，Comfy Org 团队维护）", "clearance": "REFERENCE AREA", "abstract": "节点式 AI 图像/视频生成工作流引擎，本地部署主力。", "findings": [ "模型放错目录是最常见故障，报 model not found 先查目录。" ], "source": "/docs/comfyui", "link": "#/docs/comfyui", "slot": null }, { "id": "X-022", "title": "MiniMax H3", "en": "MINIMAX H3", "department": "视频", "category": "本地部署", "date": "已上线", "lead": "闫俊杰（MiniMax 创始人、董事长兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "最强开源视频生成模型，量化后 8G 显存可跑。", "findings": [ "本地部署≠全开源：完整 2K 需官方 API。" ], "source": "/docs/minimax-h3", "link": "#/docs/minimax-h3", "slot": null }, { "id": "X-023", "title": "LLM 本地部署", "en": "LOCAL LLM", "department": "大模型", "category": "本地部署", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "Ollama 与 LM Studio 双路线入门，GGUF 量化说明。", "findings": [ "选型一句话：Ollama 吃 API 生态，LM Studio 吃图形界面（额外支持 MLX）。", "装完先改模型存储目录——默认落在 C 盘用户目录，模型动辄数个 GB。" ], "source": "/docs/llm-local", "link": "#/docs/llm-local", "slot": null }, { "id": "X-024", "title": "SD WebUI (A1111)", "en": "A1111", "department": "出图", "category": "本地部署", "date": "已上线", "lead": "AUTOMATIC1111（原作者）", "clearance": "REFERENCE AREA", "abstract": "经典出图 WebUI，扩展生态大，SD1.5/SDXL 显存门槛低。", "findings": [ "开发放缓、Flux 支持弱，追新转 Forge 或 ComfyUI。" ], "source": "/docs/a1111", "link": "#/docs/a1111", "slot": null }, { "id": "X-025", "title": "Forge", "en": "FORGE", "department": "出图", "category": "本地部署", "date": "已上线", "lead": "lllyasviel（张吕敏，作者）", "clearance": "REFERENCE AREA", "abstract": "A1111 加速分支，12G 显存跑 Flux 的舒适选择。", "findings": [ "原作者更新减缓，长期维护看社区分支 reForge / Forge Neo。" ], "source": "/docs/forge", "link": "#/docs/forge", "slot": null }, { "id": "X-026", "title": "GPT-SoVITS", "en": "GPT-SOVITS", "department": "语音", "category": "本地部署", "date": "已上线", "lead": "RVC-Boss（项目作者）", "clearance": "REFERENCE AREA", "abstract": "中文社区最热语音克隆 TTS，官方 7z 整合包解压即用。", "findings": [ "训练 batch 过高易爆显存，12G 卡从小 batch 起步。" ], "source": "/docs/gpt-sovits", "link": "#/docs/gpt-sovits", "slot": null }, { "id": "X-027", "title": "CosyVoice", "en": "COSYVOICE", "department": "语音", "category": "本地部署", "date": "已上线", "lead": "吴泳铭（阿里集团 CEO）", "clearance": "REFERENCE AREA", "abstract": "阿里多语言零样本 TTS，流式合成、指令控制。", "findings": [ "Windows 装 pynini 依赖易卡，优先整合包或 WSL。" ], "source": "/docs/cosyvoice", "link": "#/docs/cosyvoice", "slot": null }, { "id": "X-028", "title": "Fish Speech", "en": "FISH SPEECH", "department": "语音", "category": "本地部署", "date": "已上线", "lead": "Shijia Liao（Fish Audio 首席科学家）", "clearance": "REFERENCE AREA", "abstract": "开源多语言 TTS 与强克隆，情感标签、流式合成。", "findings": [ "Windows 原生支持弱，推荐 WSL / Docker。" ], "source": "/docs/fish-speech", "link": "#/docs/fish-speech", "slot": null }, { "id": "X-029", "title": "vLLM", "en": "VLLM", "department": "推理服务", "category": "本地部署", "date": "已上线", "lead": "Woosuk Kwon 等（UC Berkeley Sky Computing Lab 原创团队）", "clearance": "REFERENCE AREA", "abstract": "高吞吐 LLM 推理服务框架，生产 API 场景主力。", "findings": [ "Windows 原生不支持，官方路线是 WSL2；显存预分配要显式调低。" ], "source": "/docs/vllm", "link": "#/docs/vllm", "slot": null }, { "id": "X-030", "title": "llama.cpp", "en": "LLAMA.CPP", "department": "推理引擎", "category": "本地部署", "date": "已上线", "lead": "Georgi Gerganov（作者、首席维护者）", "clearance": "REFERENCE AREA", "abstract": "GGUF 推理引擎，全后端覆盖，Ollama 等地基。", "findings": [ "预编译包要选对显卡架构版本（CUDA / Vulkan / CPU）。" ], "source": "/docs/llama-cpp", "link": "#/docs/llama-cpp", "slot": null }, { "id": "X-031", "title": "Wan 万相", "en": "WAN", "department": "视频", "category": "本地部署", "date": "已上线", "lead": "吴泳铭（阿里集团 CEO）", "clearance": "REFERENCE AREA", "abstract": "阿里开源视频生成，小参数版消费级可跑。", "findings": [ "长视频/高分辨率显存非线性暴涨，先 480P 短片段验证。" ], "source": "/docs/wan", "link": "#/docs/wan", "slot": null }, { "id": "X-032", "title": "HunyuanVideo", "en": "HUNYUAN VIDEO", "department": "视频", "category": "本地部署", "date": "已上线", "lead": "马化腾（腾讯创始人）", "clearance": "REFERENCE AREA", "abstract": "腾讯混元视频；1.5 版消费级友好，原版 12G 别碰。", "findings": [ "认准 1.5 版，网上大量教程还是原版 13B 的配置。" ], "source": "/docs/hunyuanvideo", "link": "#/docs/hunyuanvideo", "slot": null }, { "id": "X-033", "title": "更多开源选择", "en": "MORE TOOLS", "department": "索引", "category": "本地部署", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "音乐 / 3D / 数字人 / 图像编辑等开源项目速查。", "findings": [ "索引项目迭代极快，安装与显存门槛以各自仓库 README 当期说明为准。" ], "source": "/docs/more-local", "link": "#/docs/more-local", "slot": null }, { "id": "X-034", "title": "HF 与国内镜像", "en": "HF MIRROR", "department": "知识库", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "一行环境变量切换 hf-mirror，实测 8-12MB/s。", "findings": [ "一个 HF_ENDPOINT 环境变量切镜像，huggingface-cli / transformers / diffusers 全系生效。", "镜像握手失败时切回官方源或换魔搭，两个通道都试。" ], "source": "/docs/kb/hf-mirror", "link": "#/docs/kb/hf-mirror", "slot": null }, { "id": "X-035", "title": "GGUF 量化与显存", "en": "GGUF QUANT", "department": "知识库", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "Q4_K_M 均衡之选，7-8B 模型 8G 显卡可跑。", "findings": [ "Q4_K_M 是均衡甜点；低于 Q4 质量明显下降，低于 Q3 基本不可用。", "显存表是权重占用，长上下文的 KV cache 还要额外留 1-3GB。" ], "source": "/docs/kb/gguf-quant", "link": "#/docs/kb/gguf-quant", "slot": null }, { "id": "X-036", "title": "Ollama vs LM Studio", "en": "OLLAMA vs LMS", "department": "知识库", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "命令行派与图形界面派选型对照。", "findings": [ "两者可共存互不冲突，按场景切换；同一模型两边都存会各占一份磁盘。" ], "source": "/docs/kb/ollama-vs-lmstudio", "link": "#/docs/kb/ollama-vs-lmstudio", "slot": null }, { "id": "X-037", "title": "显存与量化进阶", "en": "VRAM ADVANCED", "department": "知识库", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "显存三本账（权重/KV Cache/激活）与量化档位选择。", "findings": [ "长对话爆显存多半是 KV Cache——先降上下文再动模型。" ], "source": "/docs/kb/vram-advanced", "link": "#/docs/kb/vram-advanced", "slot": null }, { "id": "X-038", "title": "CUDA 版本匹配", "en": "CUDA MATCH", "department": "知识库", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "驱动、CUDA、PyTorch 三层链条与常见报错对照。", "findings": [ "验收标准只有一条：torch.cuda.is_available() 返回 True，别信安装日志。", "pip install torch 默认可能装到 CPU 版，须用官网选择器生成的 index-url 装对应轮子。" ], "source": "/docs/kb/cuda-match", "link": "#/docs/kb/cuda-match", "slot": null }, { "id": "X-039", "title": "视频生成调参", "en": "VIDEO TUNING", "department": "知识库", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "分辨率与帧数是乘法关系，别两个一起拉。", "findings": [ "显存不够按量化、offload、降参数的顺序处理，别硬撑。" ], "source": "/docs/kb/video-tuning", "link": "#/docs/kb/video-tuning", "slot": null }, { "id": "X-040", "title": "LoRA 训练入门", "en": "LORA BASICS", "department": "知识库", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "低秩微调的门槛、数据集与过拟合判断。", "findings": [ "数据质量优先于数量；同一主体要覆盖角度/光线/背景的多样性。", "过拟合典型信号=画面只认训练集构图，宁可少步数或提前停。" ], "source": "/docs/kb/lora-basics", "link": "#/docs/kb/lora-basics", "slot": null }, { "id": "X-041", "title": "模型权重直达", "en": "MODEL HUB", "department": "下载", "category": "知识库与索引", "date": "已上线", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "混元 / 千问 / DeepSeek 官方与镜像地址汇总。", "findings": [ "先分清原始权重与 GGUF 量化版：本地推理选 GGUF，训练与微调用原始权重。", "GitHub 组织改名会导致旧链接失效（如 THUDM→zai-org），以官网公告为准。" ], "source": "/docs/models-hub", "link": "#/docs/models-hub", "slot": null }, { "id": "X-042", "title": "站长介绍", "en": "ABOUT", "department": "站点", "category": "站点与自研", "date": "已上线", "lead": "Admin（站长）", "clearance": "REFERENCE AREA", "abstract": "站长与这个站的来龙去脉。", "findings": [ "本页是全站唯一带站长个人色彩的内容块，文案定稿后替换占位。" ], "source": "/about", "link": "#/about", "slot": null }, { "id": "X-043", "title": "作品展示", "en": "WORKS", "department": "站点", "category": "站点与自研", "date": "已上线", "lead": "Admin（站长）", "clearance": "REFERENCE AREA", "abstract": "当前展示：AstrBot 开源项目。", "findings": [ "分「自研项目」（DanTide 弹潮 / 轻舟 Agent）与「开源项目」（AstrBot）两区。" ], "source": "/works", "link": "#/works", "slot": null }, { "id": "X-044", "title": "大模型实时排名", "en": "MODEL RANK", "department": "情报", "category": "站点与自研", "date": "已上线", "lead": "Admin（站长）", "clearance": "REFERENCE AREA", "abstract": "多源榜单投票制综合，脚本抓取，无 LLM 参与。", "findings": [ "数据来自 LMArena 四分类与 Artificial Analysis，Workers 定时缓存。" ], "source": "/rankings", "link": "#/rankings", "slot": null }, { "id": "X-045", "title": "DeepSeek Harness", "en": "DSH", "department": "自研", "category": "站点与自研", "date": "已上线", "lead": "梁文锋（DeepSeek 创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "自研 DeepSeek 定制 harness 整合包（筹备中）。", "findings": [ "整合包绝不预置 API key，密钥首次运行现场填写。" ], "source": "/docs/dsh", "link": "#/docs/dsh", "slot": null }, { "id": "X-046", "title": "速度测试", "en": "SPEEDTEST", "department": "站点功能", "category": "站点与自研", "date": "已上线", "lead": "Admin（站长）", "clearance": "REFERENCE AREA", "abstract": "浏览器内实测延迟/抖动/下载带宽，数据不出站；后端子域部署后自动切换正式通道。", "findings": [ "本地基准模式用同源资源吞吐 + 浏览器 RTT 估算，非真实带宽上限。" ], "source": "/speedtest", "link": "#/speedtest", "slot": null }, { "id": "X-047", "title": "键盘快捷键", "en": "SHORTCUTS", "department": "效率", "category": "站点与自研", "date": "已上线", "lead": "Admin（站长）", "clearance": "REFERENCE AREA", "abstract": "命令面板（Ctrl+K）、档案墙翻档、全站键盘操作与无障碍适配速查。", "findings": [ "全站 Tab 可达且有焦点圈；系统“减少动态效果”开启时全部动效静止。" ], "source": "/docs/shortcuts", "link": "#/docs/shortcuts", "slot": null }, { "id": "X-048", "title": "更新日志", "en": "CHANGELOG", "department": "站点功能", "category": "站点与自研", "date": "已上线", "lead": "Admin（站长）", "clearance": "REFERENCE AREA", "abstract": "按版本记录每次更新：新增 / 调整 / 修复三组，与代码提交一一对应。", "findings": [ "新功能上线先看这里；站点是活的，不是一次性作品。" ], "source": "/changelog", "link": "#/changelog", "slot": null }, { "id": "X-049", "title": "DeepSeek 深度求索", "en": "DEEPSEEK", "department": "厂商 · 权重开放", "category": "厂商·国产", "date": "已上线", "lead": "梁文锋（创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "国产开源旗舰：V 系列与 R1 蒸馏版大量开源，MIT 许可，本地与 API 双线可用。", "findings": [ "权重开放度国产第一梯队，Ollama / LM Studio / vLLM 都有一条命令的路径。", "蒸馏小模型能塞进消费级显卡，大尺寸仍要多卡或大内存工作站。", "API 走 Anthropic 兼容端点接入各编程客户端，端点路径按官方文档填。", "中文与代码能力均衡，长上下文场景先看显存账再决定档位。", "避坑：同名系列的量化档差别很大，认 Q4_K_M 以上的成熟档位，别贪体积。" ], "source": "/docs/vendors/deepseek", "link": "#/docs/vendors/deepseek", "slot": null }, { "id": "X-050", "title": "阿里云 Qwen 通义千问", "en": "QWEN / ALIBABA", "department": "厂商 · 双线并行", "category": "厂商·国产", "date": "已上线", "lead": "吴泳铭（集团 CEO）、周靖人（阿里云 CTO）、林俊旸（前 Qwen 负责人）", "clearance": "REFERENCE AREA", "abstract": "开源生态最完整的国产厂商：中小尺寸 Apache 2.0 开源，旗舰档走云 API。", "findings": [ "下载量与衍生模型数量全球前列，微调与本地基础模型的首选之一。", "中小尺寸开源、Max/Pro 档闭源 API——同一品牌两条线，选型先确认档位。", "Ollama / LM Studio / vLLM 全支持，多模态与多语言覆盖广。", "衍生微调模型极多，找社区版本时看清底座尺寸与许可。", "避坑：不同尺寸的许可不完全一致（部分 Apache 2.0、更大的另有条款），商用逐个确认。" ], "source": "/docs/vendors/qwen", "link": "#/docs/vendors/qwen", "slot": null }, { "id": "X-051", "title": "月之暗面 Kimi", "en": "MOONSHOT KIMI", "department": "厂商 · 长上下文", "category": "厂商·国产", "date": "已上线", "lead": "杨植麟（创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "以长上下文与代理能力见长；K 系列部分开放权重，官方 CLI 单文件即可运行。", "findings": [ "长上下文是主打卖点，适合长文档与多步代理任务。", "官方 CLI 走 code.kimi.com 官方脚本，npm 上的同名包是无关抢注包。", "开放权重部分可用 Ollama 与社区工具加载，旗舰能力仍以官方服务为准。", "编程套餐与网页端额度分开计算，团队用先理清计费口径。", "避坑：认准官方域名下载，第三方加速版可能夹带改动。" ], "source": "/docs/vendors/moonshot", "link": "#/docs/vendors/moonshot", "slot": null }, { "id": "X-052", "title": "智谱 GLM", "en": "ZHIPU GLM", "department": "厂商 · 权重开放", "category": "厂商·国产", "date": "已上线", "lead": "张鹏（CEO）、唐杰（创始人）、刘德兵（董事长）", "clearance": "REFERENCE AREA", "abstract": "MIT 许可的开源线加上企业服务：编码与长程代理能力在独立评测中常居开源前列。", "findings": [ "GLM 系列以 MIT 许可开源，Ollama 与主流运行器均支持。", "编程套餐兼容 Claude Code 接入方式，接入时 base url 要用官方专用路径。", "企业侧有国产硬件适配，合规与本地化部署友好。", "集团图标与 CodeGeeX 同源，档案图标沿用同一枚（来源清单有标注）。", "避坑：Anthropic 兼容接入的路径随手改会 404，照官方文档抄。" ], "source": "/docs/vendors/zhipu", "link": "#/docs/vendors/zhipu", "slot": null }, { "id": "X-053", "title": "字节跳动 豆包 / Seed", "en": "BYTEDANCE DOUBAO", "department": "厂商 · 闭源为主", "category": "厂商·国产", "date": "已上线", "lead": "梁汝波（CEO）、吴永辉（Seed 基础研究负责人）", "clearance": "REFERENCE AREA", "abstract": "消费级规模最大的国产平台之一：豆包 App 加火山引擎 API，旗舰以闭源为主。", "findings": [ "平台分发极强，内容与推荐生态结合紧，价格有竞争力。", "闭源旗舰走火山引擎 API 与豆包端，权重不对外提供。", "Seed 系列部分能力对外以服务形式开放，本地化不在其路线内。", "已有火山引擎体系的团队接入最省事。", "避坑：先想清要不要本地跑——要本地就得换开源线厂商。" ], "source": "/docs/vendors/bytedance", "link": "#/docs/vendors/bytedance", "slot": null }, { "id": "X-054", "title": "百度 文心 ERNIE", "en": "BAIDU ERNIE", "department": "厂商 · 闭源为主", "category": "厂商·国产", "date": "已上线", "lead": "李彦宏（创始人、董事长兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "搜索与知识增强见长；旗舰闭源走千帆平台，部分小模型已开放权重。", "findings": [ "中文场景成熟，政务与企业采用面广。", "千帆平台提供 API 与企业交付，旗舰不开放权重。", "部分小尺寸模型已开源，可在本地跑通流程再决定是否上云。", "与百度智能云绑定的行业方案较多。", "避坑：同一品牌下开源与闭源档位分开看，别把部分开源当成全量开源。" ], "source": "/docs/vendors/baidu", "link": "#/docs/vendors/baidu", "slot": null }, { "id": "X-055", "title": "腾讯 混元 Hunyuan", "en": "TENCENT HUNYUAN", "department": "厂商 · 双线并行", "category": "厂商·国产", "date": "已上线", "lead": "马化腾（创始人）、姚顺雨（首席 AI 科学家）", "clearance": "REFERENCE AREA", "abstract": "社交与协作生态内嵌：混元部分型号开源，旗舰走腾讯云与微信、企业微信侧。", "findings": [ "企业协作场景优势明显，生态整合好。", "部分型号开放权重，旗舰偏闭源 API。", "云侧走腾讯云，端侧出现在微信、企业微信与元宝。", "已用腾讯云的企业接入成本最低。", "避坑：开源型号与旗舰能力差距明显，选型按任务而不是按品牌。" ], "source": "/docs/vendors/tencent", "link": "#/docs/vendors/tencent", "slot": null }, { "id": "X-056", "title": "华为 盘古 Pangu", "en": "HUAWEI PANGU", "department": "厂商 · 私有化交付", "category": "厂商·国产", "date": "已上线", "lead": "任正非（创始人）、余承东（盘古团队负责人）", "clearance": "REFERENCE AREA", "abstract": "自主可控路线的代表：盘古系列配合昇腾硬件，以政企私有化交付为主。", "findings": [ "国产硬件绑定强，昇腾适配是核心卖点。", "权重不公开，交付形态是私有化与行业解决方案。", "政企与自主可控场景的首选之一。", "消费级本地部署不在其路线内，个人开发者可直接跳过。", "避坑：私有化交付通常需要商务合同与实施周期，不是下载就能用。" ], "source": "/docs/vendors/huawei", "link": "#/docs/vendors/huawei", "slot": null }, { "id": "X-057", "title": "科大讯飞 星火", "en": "IFLYTEK SPARK", "department": "厂商 · 行业方案", "category": "厂商·国产", "date": "已上线", "lead": "刘庆峰（创始人、董事长兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "语音与教育行业深耕：星火系列走开放平台 API 与行业解决方案。", "findings": [ "语音识别与合成是传统强项，教育场景积累深。", "闭源为主，行业方案多，个人开发走开放平台 API。", "讯飞组织头像已用于站内 iFlyCode 档案，本次沿用同一枚图标。", "如需语音链路，可与本地开源语音模型（GPT-SoVITS / CosyVoice）对照选型。", "避坑：行业方案的计费与配额差异大，接入前先要测试额度。" ], "source": "/docs/vendors/iflytek", "link": "#/docs/vendors/iflytek", "slot": null }, { "id": "X-058", "title": "MiniMax", "en": "MINIMAX", "department": "厂商 · 双线并行", "category": "厂商·国产", "date": "已上线", "lead": "闫俊杰（创始人、董事长兼 CEO 兼 CTO）", "clearance": "REFERENCE AREA", "abstract": "多模态与性价比路线：部分模型开放权重，旗舰与 H 系列走平台服务。", "findings": [ "多模态与消费级应用落地多，性价比好。", "开放程度低于 DeepSeek 与 Qwen，本地可选面较窄。", "开源部分可用社区工具加载，旗舰能力以官方服务为准。", "站内已有 MiniMax Code CLI 与 H3 两条工具档案，厂商档案作为上层索引。", "避坑：先确认目标模型到底在不在开源列表里。" ], "source": "/docs/vendors/minimax", "link": "#/docs/vendors/minimax", "slot": null }, { "id": "X-059", "title": "商汤 SenseTime", "en": "SENSETIME", "department": "厂商 · 行业方案", "category": "厂商·国产", "date": "已上线", "lead": "徐立（联合创始人、执行主席兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "视觉与行业 AI 积累深：日日新系列走平台 API 与私有化部署。", "findings": [ "CV 起家的行业 AI 玩家，视觉与多模态积累深。", "闭源为主，交付形态是平台 API 与私有化部署。", "政务、金融、医疗等行业解决方案是主要落地方式。", "个人开发者较少直接使用，可关注其开源的多模态成果。", "避坑：行业项目通常需要商务与实施周期，不是自助开通。" ], "source": "/docs/vendors/sensetime", "link": "#/docs/vendors/sensetime", "slot": null }, { "id": "X-060", "title": "小米 MiMo", "en": "XIAOMI MIMO", "department": "厂商 · 权重开放", "category": "厂商·国产", "date": "已上线", "lead": "雷军（创始人）、罗福莉（MiMo 负责人）", "clearance": "REFERENCE AREA", "abstract": "手机厂商里开源意识最强的一家：MiMo 系列部分开放权重，路由与编码流量表现突出。", "findings": [ "在部分路由与编码流量统计里表现显眼。", "有开放权重选项，可配合社区工具本地跑。", "与小米终端生态（手机、汽车、家居）绑定深。", "适合想试国产小尺寸模型的开发者做对照。", "避坑：开放的多是小尺寸档，能力与云端旗舰差距明显。" ], "source": "/docs/vendors/xiaomi", "link": "#/docs/vendors/xiaomi", "slot": null }, { "id": "X-061", "title": "阶跃星辰 StepFun", "en": "STEPFUN", "department": "厂商 · 闭源为主", "category": "厂商·国产", "date": "已上线", "lead": "姜大昕（创始人兼 CEO）、印奇（董事长）", "clearance": "REFERENCE AREA", "abstract": "多模态与终端合作路线：Step 系列走平台服务，部分小模型开放。", "findings": [ "多模态（图、音、视频）是其对外重点。", "旗舰闭源走平台 API，部分小尺寸开放权重。", "与国产终端厂商的预装合作较多。", "开发者侧可从小尺寸开源档切入试水。", "避坑：不同档位的开放程度不一，选型前先确认权重可不可得。" ], "source": "/docs/vendors/stepfun", "link": "#/docs/vendors/stepfun", "slot": null }, { "id": "X-062", "title": "零一万物 01.AI", "en": "01.AI", "department": "厂商 · 双线并行", "category": "厂商·国产", "date": "已上线", "lead": "李开复（创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "Yi 系列早期以开源著称，如今开源与闭源 API 并行。", "findings": [ "早期 Yi 系列开源知名度高，社区衍生版本多。", "现有开源线（小尺寸）与闭源 API 并存。", "偏企业落地与行业方案。", "可直接在本地跑其开源档做中文对照实验。", "避坑：开源线与闭源线的能力与许可口径不同，别混用宣传页数据。" ], "source": "/docs/vendors/01ai", "link": "#/docs/vendors/01ai", "slot": null }, { "id": "X-063", "title": "OpenAI", "en": "OPENAI", "department": "厂商 · 闭源云", "category": "厂商·美国", "date": "已上线", "lead": "Sam Altman（CEO）、Greg Brockman（总裁）", "clearance": "REFERENCE AREA", "abstract": "消费级规模最大的前沿实验室：GPT 系列加 ChatGPT，API 与 Azure 双通道。", "findings": [ "消费级用户规模最大，编码与通用能力长期在第一梯队。", "权重不开放；gpt-oss 属有限开源尝试，能力与旗舰有差距。", "交付走官方 API、ChatGPT 端、Azure OpenAI 三种通道。", "已有 Azure 体系的企业走 Azure 通道，合规条款更完整。", "避坑：桌面与网页客户端仍是云端推理，断网不可用；本地跑 ChatGPT 不存在。" ], "source": "/docs/vendors/openai", "link": "#/docs/vendors/openai", "slot": null }, { "id": "X-064", "title": "Anthropic", "en": "ANTHROPIC", "department": "厂商 · 闭源云", "category": "厂商·美国", "date": "已上线", "lead": "Dario Amodei（联合创始人兼 CEO）、Jack Clark（联合创始人）", "clearance": "REFERENCE AREA", "abstract": "企业编码与复杂推理的领先者：Claude 系列加 Claude Code，安全对齐口碑好。", "findings": [ "企业 LLM 支出份额与编码场景口碑领先。", "完全闭源；权重不可得，本地部署不在其路线。", "官方 API、Claude 端、AWS Bedrock 三条通道。", "站内 Claude Code、CC Switch 等档案就是围绕它的接入生态。", "避坑：品牌橙色属于 Claude 标识，Anthropic 母品牌是单色标志——档案图标用的是后者。" ], "source": "/docs/vendors/anthropic", "link": "#/docs/vendors/anthropic", "slot": null }, { "id": "X-065", "title": "Google DeepMind", "en": "GOOGLE DEEPMIND", "department": "厂商 · 双线并行", "category": "厂商·美国", "date": "已上线", "lead": "Demis Hassabis（联合创始人兼 CEO）、Koray Kavukcuoglu（SVP）", "clearance": "REFERENCE AREA", "abstract": "多模态与长上下文见长：Gemini 旗舰闭源，Gemma 系列按 Apache 2.0 开源。", "findings": [ "与搜索、办公、Android 生态深度绑定，多模态能力强。", "Gemini 旗舰闭源（API / Vertex AI / Gemini App），Gemma 开源可本地跑。", "Gemma 轻量版适合边缘与笔记本，大尺寸在单张高端卡上可流畅运行。", "云侧走 Google Cloud Vertex AI，端侧出现在 Workspace。", "避坑：Gemini 与 Gemma 是两条完全不同的产品线，别把两者混为一谈。" ], "source": "/docs/vendors/google", "link": "#/docs/vendors/google", "slot": null }, { "id": "X-066", "title": "Meta", "en": "META", "department": "厂商 · 双线并行", "category": "厂商·美国", "date": "已上线", "lead": "Mark Zuckerberg（CEO）、Alexandr Wang（首席 AI 官，领导 MSL）", "clearance": "REFERENCE AREA", "abstract": "开源权重生态的奠基者：Llama 系列开源（社区许可），闭源线走 Meta AI 的 Muse。", "findings": [ "生态最成熟：教程、量化版本、微调工具几乎都优先适配 Llama。", "Llama 需接受社区许可后下载权重；Muse 闭源只出现在 Meta 应用内。", "Ollama / LM Studio / llama.cpp / Transformers 全支持本地部署。", "消费级分发极强（Instagram / WhatsApp），闭源线注重应用体验。", "避坑：许可不是标准开源协议，商用前按条款逐条确认（尤其超大参数档）。" ], "source": "/docs/vendors/meta", "link": "#/docs/vendors/meta", "slot": null }, { "id": "X-067", "title": "Microsoft", "en": "MICROSOFT", "department": "厂商 · 双线并行", "category": "厂商·美国", "date": "已上线", "lead": "萨提亚·纳德拉（CEO）、Mustafa Suleyman（Microsoft AI CEO）", "clearance": "REFERENCE AREA", "abstract": "企业办公与合规路线：Phi 小模型按 MIT 类许可开源，旗舰能力走 Azure OpenAI。", "findings": [ "小模型推理与 STEM 能力出色，笔记本与低配设备友好。", "Phi 系列开源（MIT 类），Ollama 与 LM Studio 直接支持。", "企业侧核心旗舰依赖云（Azure AI 与 Copilot），并非本地推理。", "与 Office、Entra 生态绑定紧，合规与审计能力强。", "避坑：Copilot 是云服务入口，不是本地模型；要离线只能走 Phi 这类开源线。" ], "source": "/docs/vendors/microsoft", "link": "#/docs/vendors/microsoft", "slot": null }, { "id": "X-068", "title": "Nvidia", "en": "NVIDIA", "department": "厂商 · 权重开放", "category": "厂商·美国", "date": "已上线", "lead": "黄仁勋（创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "硬件与推理栈的制定者：Nemotron 系列开放权重，配 TensorRT-LLM 或 vLLM 追求吞吐。", "findings": [ "推理效率极高，与自家硬件强绑定。", "Nemotron 权重开放，Ollama 等运行器也支持。", "生产环境常配 TensorRT-LLM 或 vLLM，高吞吐场景优势明显。", "显存与量化账见站内知识库，选卡先看显存而不是型号。", "避坑：硬件绑定强，非 N 卡环境收益会明显下降。" ], "source": "/docs/vendors/nvidia", "link": "#/docs/vendors/nvidia", "slot": null }, { "id": "X-069", "title": "Amazon", "en": "AMAZON", "department": "厂商 · 闭源云", "category": "厂商·美国", "date": "已上线", "lead": "Andy Jassy（CEO）", "clearance": "REFERENCE AREA", "abstract": "云市场托管路线：Nova 与 Titan 系列经 Amazon Bedrock 提供，服务已有 AWS 客户。", "findings": [ "已有 AWS 体系的企业接入最省事，模型广度好。", "闭源托管，权重不开放；Bedrock 上仍是托管推理。", "与其他云市场（Vertex / Azure）选任一即可，别为多云加一层复杂度。", "常与 S3、Lambda 等既有服务串起来做 RAG 与批处理。", "避坑：上了 Bedrock 不等于能离线跑——断网即不可用。" ], "source": "/docs/vendors/amazon", "link": "#/docs/vendors/amazon", "slot": null }, { "id": "X-070", "title": "xAI", "en": "XAI", "department": "厂商 · 闭源云", "category": "厂商·美国", "date": "已上线", "lead": "Elon Musk（创始人，2026-07 并入 SpaceX）", "clearance": "REFERENCE AREA", "abstract": "实时知识与风格独特：Grok 系列经官方 API 与 X 平台分发。", "findings": [ "与 X 平台数据结合，实时性是其定位。", "完全闭源，权重不可得。", "交付走官方 API 与 Grok 应用，规模仍落后于前三家。", "接入方式与主流 OpenAI 兼容接口接近，迁移成本低。", "避坑：实时数据是双刃剑，事实核查流程要自己补。" ], "source": "/docs/vendors/xai", "link": "#/docs/vendors/xai", "slot": null }, { "id": "X-071", "title": "Cohere", "en": "COHERE", "department": "厂商 · 企业检索", "category": "厂商·美国", "date": "已上线", "lead": "Aidan Gomez（联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "企业检索增强（RAG）路线：Command 系列加检索与重排工具链，云中立部署。", "findings": [ "企业检索增强是其主战场，重排能力常被单独选型。", "闭源为主，权重不开放。", "可部署在多家云或私有环境，云中立性是其卖点。", "常与向量库、重排模型组合成完整 RAG 链路。", "避坑：单看生成质量不突出，价值在整条检索链路的配套。" ], "source": "/docs/vendors/cohere", "link": "#/docs/vendors/cohere", "slot": null }, { "id": "X-072", "title": "Perplexity", "en": "PERPLEXITY", "department": "厂商 · 搜索增强", "category": "厂商·美国", "date": "已上线", "lead": "Aravind Srinivas（联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "搜索增强路线：在线答案引擎加 Sonar 系列 API，主打可引用来源。", "findings": [ "以带引用的答案见长，检索体验是其差异点。", "闭源，权重不开放。", "API 走 OpenAI 兼容形态，接入成本低。", "适合做需要来源可核查的检索类应用。", "避坑：引用不等于核实，关键结论仍要自己打开源链接确认。" ], "source": "/docs/vendors/perplexity", "link": "#/docs/vendors/perplexity", "slot": null }, { "id": "X-073", "title": "Thinking Machines Lab", "en": "THINKING MACHINES", "department": "厂商 · 权重开放", "category": "厂商·美国", "date": "已上线", "lead": "Mira Murati（创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "新锐实验室：Inkling 等开放权重模型被视为当前较强的美国开源选项。", "findings": [ "公开权重、许可较宽松，是美系开源的新变量。", "体量小、迭代快，社区跟进积极。", "可通过 Hugging Face 与主流运行器加载。", "适合做「美系开源 vs 国产开源」的横向对照。", "避坑：初创机构的长期维护能力要打问号，别把关键业务单点押上。" ], "source": "/docs/vendors/thinking-machines", "link": "#/docs/vendors/thinking-machines", "slot": null }, { "id": "X-074", "title": "Arcee AI", "en": "ARCEE AI", "department": "厂商 · 权重开放", "category": "厂商·美国", "date": "已上线", "lead": "Mark McQuade（联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "垂直与小尺寸开源路线：面向特定合规与私有化场景。", "findings": [ "以开源、半开源的小尺寸模型为主，许可相对宽松。", "定位细分：为自有硬件或行业场景做定制。", "模型卡与量化版本在社区可获取。", "适合需要可控小模型的本地场景。", "避坑：生态与文档规模远小于头部，排错要靠自己。" ], "source": "/docs/vendors/arcee", "link": "#/docs/vendors/arcee", "slot": null }, { "id": "X-075", "title": "Poolside", "en": "POOLSIDE", "department": "厂商 · 权重开放", "category": "厂商·美国", "date": "已上线", "lead": "Jason Warner（联合创始人兼 CEO）、Eiso Kant（联合创始人兼 CTO）", "clearance": "REFERENCE AREA", "abstract": "代码生成方向的专精玩家：部分模型开放权重，主打企业私有化。", "findings": [ "专注代码生成，不做通用全能路线。", "部分模型开放权重，另有企业版交付。", "面向有代码隐私要求的企业场景。", "可与本地代码模型（Qwen-Coder 类）横向对比。", "避坑：开源档通常落后其商用档一代，评测数据要看对应版本。" ], "source": "/docs/vendors/poolside", "link": "#/docs/vendors/poolside", "slot": null }, { "id": "X-076", "title": "Mistral AI", "en": "MISTRAL AI", "department": "厂商 · 双线并行", "category": "厂商·欧洲", "date": "已上线", "lead": "Arthur Mensch（联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "欧洲综合最强：开源权重（Apache 2.0）与闭源旗舰并行，主权部署友好。", "findings": [ "小模型边缘友好、许可宽松，性能接近全球前列。", "Ministral 等开源档可本地跑，旗舰档以 API 优先。", "支持欧洲境内基础设施与本地选项，数据不出境的卖点明确。", "非美中替代的首选，文档与工具支持完整。", "避坑：开源与闭源档分开看，商用前逐个确认许可与数据条款。" ], "source": "/docs/vendors/mistral-ai", "link": "#/docs/vendors/mistral-ai", "slot": null }, { "id": "X-077", "title": "Aleph Alpha", "en": "ALEPH ALPHA", "department": "厂商 · 主权交付", "category": "厂商·欧洲", "date": "已上线", "lead": "—（公开信息显示正与 Cohere 合并，负责人待官方确认）", "clearance": "REFERENCE AREA", "abstract": "德国主权 AI 代表：Pharia 系列面向受监管行业，on-prem 与欧洲基础设施交付。", "findings": [ "合规与审计能力强，政务与受监管企业反馈积极。", "偏专有与闭源交付，公开消费级权重很少。", "交付形态是私有化与欧洲基础设施，不是下载即用。", "适合数据不能出境的行业场景。", "避坑：普通开发者本地使用门槛高，个人项目可先跳过。" ], "source": "/docs/vendors/aleph-alpha", "link": "#/docs/vendors/aleph-alpha", "slot": null }, { "id": "X-078", "title": "DeepL", "en": "DEEPL", "department": "厂商 · 翻译专用", "category": "厂商·欧洲", "date": "已上线", "lead": "Jarek Kutylowski（联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "专业翻译的事实标准：翻译与写作模型经 API 与网页、桌面服务提供。", "findings": [ "欧洲语言互译质量长期领先，术语一致性处理成熟。", "闭源，权重不开放，无本地部署路径。", "API 集成简单，适合嵌进文档与本地化流程。", "术语表（Glossary）功能是专业场景的关键。", "避坑：要离线翻译得看本地开源翻译模型，DeepL 无法下沉到本机。" ], "source": "/docs/vendors/deepl", "link": "#/docs/vendors/deepl", "slot": null }, { "id": "X-079", "title": "LightOn", "en": "LIGHTON", "department": "厂商 · 气隙部署", "category": "厂商·欧洲", "date": "已上线", "lead": "Igor Carron、Laurent Daudet（联席 CEO）", "clearance": "REFERENCE AREA", "abstract": "文档智能与主权部署：支持 on-prem 与气隙（air-gapped）环境。", "findings": [ "长文档与 RAG 场景见长。", "偏专有交付，气隙环境支持是其差异点。", "面向对数据边界要求最严的客户（政企、国防相关）。", "个人开发者可以用站内开源文档智能链路替代。", "避坑：气隙交付通常需定制与实施周期，不适合快速试验。" ], "source": "/docs/vendors/lighton", "link": "#/docs/vendors/lighton", "slot": null }, { "id": "X-080", "title": "Kyutai", "en": "KYUTAI", "department": "厂商 · 研究开源", "category": "厂商·欧洲", "date": "已上线", "lead": "Patrick Pérez（CEO）", "clearance": "REFERENCE AREA", "abstract": "法国研究型机构：语音与文本方向的开源尝试，偏研究而非产品化。", "findings": [ "语音交互（实时语音对话）是其公开方向。", "成果多以研究与开源形式发布，许可随项目而异。", "适合关注语音链路前沿的人跟踪，不适合直接上生产。", "与站内本地语音工具（GPT-SoVITS / CosyVoice）互为参考。", "避坑：研究项目迭代与维护节奏不稳定，别押在生产依赖上。" ], "source": "/docs/vendors/kyutai", "link": "#/docs/vendors/kyutai", "slot": null }, { "id": "X-081", "title": "H Company", "en": "H COMPANY", "department": "厂商 · 计算机代理", "category": "厂商·欧洲", "date": "已上线", "lead": "Gautier Cloix（总裁兼 CEO，2025-06 起）", "clearance": "REFERENCE AREA", "abstract": "法国方向：计算机使用代理（computer-use agents），专有交付。", "findings": [ "聚焦「代理操作电脑」这条赛道，不是通用聊天模型。", "闭源、专有交付，面向企业自动化场景。", "与欧洲主权 AI 生态关系紧密。", "研究性质较强，适合关注代理前沿的人跟踪。", "避坑：代理类产品落地依赖具体流程适配，别指望开箱即用。" ], "source": "/docs/vendors/h-company", "link": "#/docs/vendors/h-company", "slot": null }, { "id": "X-082", "title": "Helsing", "en": "HELSING", "department": "厂商 · 国防专用", "category": "厂商·欧洲", "date": "已上线", "lead": "Torsten Reil、Gundbert Scherf（联席 CEO）", "clearance": "REFERENCE AREA", "abstract": "欧洲防务 AI 代表：面向国防与安全领域，严格专有交付。", "findings": [ "客户是国防与安全机构，公开信息有限。", "严格专有，无公开权重与自助 API。", "欧洲防务自主化政策的直接受益者。", "普通开发者与民用项目基本接触不到。", "避坑：这类厂商的公开评测几乎不存在，别把营销材料当性能证据。" ], "source": "/docs/vendors/helsing", "link": "#/docs/vendors/helsing", "slot": null }, { "id": "X-083", "title": "Claude 订阅档", "en": "CLAUDE PLANS", "department": "套餐 · 国际", "category": "编程套餐", "date": "2026-09", "lead": "Dario Amodei（Anthropic 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "Claude Code 背后的账号订阅：Pro / Max 两档按倍率给用量，团队按座位计费。", "findings": [ "定价：Pro ≈$20/月（年付 ≈$17）；Max 5x ≈$100/月；Max 20x ≈$200/月；Team / Enterprise 按座位另计。", "涉及模型：Claude 系列（旗舰与轻量档）。", "用量机制：5 小时滚动窗口 + 周限额，超额走 API 按 token 付费。", "配套：Claude Code CLI、Claude 网页与桌面端、CC Switch 切换供应商。", "避坑：窗口严格，重度重构容易在窗口内被限流——先估算 token 再选档。" ], "source": "https://claude.com/pricing", "link": "#/docs/claude-code", "slot": null }, { "id": "X-084", "title": "Cursor 套餐", "en": "CURSOR PLANS", "department": "套餐 · 国际", "category": "编程套餐", "date": "2026-09", "lead": "Michael Truell（Anysphere 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "AI 原生 IDE 的用量池订阅：Hobby 免费起步，Pro / Pro+ / Ultra 按倍数扩容。", "findings": [ "定价：Hobby 免费（有限 Agent 请求与补全）；Pro ≈$20/月（含约 $20 等值用量）；Pro+ ≈$60/月（3 倍）；Ultra ≈$200/月（20 倍）；Teams ≈$40/座位/月。", "涉及模型：Claude、GPT、Gemini 等多家可切换。", "计费形态：固定订阅正在向用量池迁移，超出按量。", "配套：自有 IDE 内的 Agent 模式，重度用户常直接升 Ultra。", "避坑：先算清每月用量池够不够，别按标价档位倒推。" ], "source": "https://cursor.com/pricing", "link": null, "slot": null }, { "id": "X-085", "title": "GitHub Copilot 套餐", "en": "COPILOT PLANS", "department": "套餐 · 国际", "category": "编程套餐", "date": "2026-09", "lead": "萨提亚·纳德拉（Microsoft CEO）", "clearance": "REFERENCE AREA", "abstract": "门槛最低的 AI 编程订阅：补全近乎不限，Agent 用量改用 Credits 计量。", "findings": [ "定价：Free（有限）；Pro ≈$10/月；Pro+ ≈$39/月；Max ≈$100/月；Business ≈$19/座位/月，Enterprise 更高。", "涉及模型：多家（含 Anthropic / OpenAI / Google 模型）。", "计费形态：2026 起转为 AI Credits，代码补全通常不限量。", "配套：VS Code、JetBrains、Visual Studio 插件，与 GitHub 深度集成。", "避坑：Agent 任务按 Credits 扣，跑长任务前先看剩余额度。" ], "source": "https://github.com/features/copilot/plans", "link": null, "slot": null }, { "id": "X-086", "title": "Codex / ChatGPT 套餐", "en": "CODEX PLANS", "department": "套餐 · 国际", "category": "编程套餐", "date": "2026-09", "lead": "Sam Altman（OpenAI CEO）", "clearance": "REFERENCE AREA", "abstract": "Codex CLI 与 ChatGPT 共用一套账号额度，Plus 起步即可跑代理编程。", "findings": [ "定价：Free / Go ≈$0–8/月（有限）；Plus ≈$20/月；Pro 5x / 20x ≈$100–200/月；Business 按座位。", "涉及模型：GPT 系列。", "形态：Codex CLI 轻量、与 GPT 生态结合好；也可改走 API 按 token 计费。", "共享机制：CLI 与网页端共用额度。", "避坑：长任务前先确认剩余额度，共享池消耗比想象快。" ], "source": "https://openai.com/chatgpt/pricing", "link": "#/docs/codex", "slot": null }, { "id": "X-087", "title": "OpenCode 套餐", "en": "OPENCODE PLANS", "department": "套餐 · 开源", "category": "编程套餐", "date": "2026-09", "lead": "Jay V、Frank Wang（SST 团队）", "clearance": "REFERENCE AREA", "abstract": "开源终端代理的托管套餐：核心永久免费（自带 Key），Go / Zen 走托管模型。", "findings": [ "定价：核心免费（BYOK）；Go 首月 ≈$5、之后 ≈$10/月（含 5 小时 / 周 / 月三层限额，如 $12 / 5 小时、$60 / 月）；Zen 按量预付约 $20 起；Black 系列 $20 / $100 / $200 倍率档；Enterprise 定制。", "涉及模型：Go 托管 DeepSeek、GLM、Kimi、Qwen、MiniMax 等开源档；Zen 接前沿闭源模型；BYOK 可接任意模型（含本地 Ollama）。", "集成：`ollama launch opencode` 一键用本地模型驱动。", "适合：隐私优先、多模型切换、成本敏感。", "避坑：托管档三层限额叠加，按自己的用量节奏选档而不是看价格。" ], "source": "https://opencode.ai", "link": "#/docs/opencode", "slot": null }, { "id": "X-088", "title": "Ollama（本地 + Cloud）", "en": "OLLAMA PLANS", "department": "套餐 · 本地优先", "category": "编程套餐", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "本地推理永久免费；Cloud 提供托管额度，一条 launch 命令把本地模型接进编程工具。", "findings": [ "定价：本地免费；Cloud 有 Free、Pro（≈$20/月含 ≈$60 额度）、Max（≈$100/月含 ≈$300 额度）。", "关键命令：`ollama launch claude` / `opencode` / `codex` / `droid`，以及 Goose、OpenClaw 等助手。", "推荐编码模型：qwen3-coder 系列、GLM 系、gpt-oss、DeepSeek 编码变体；上下文建议 64k 以上。", "编辑器集成：VS Code、Cline、JetBrains、Zed、Xcode。", "避坑：本地跑要留足显存与上下文（64k 起），量化档位见知识库两篇。" ], "source": "https://ollama.com/pricing", "link": "#/docs/kb/ollama-vs-lmstudio", "slot": null }, { "id": "X-089", "title": "Devin Desktop（原 Windsurf）", "en": "DEVIN DESKTOP", "department": "套餐 · 国际", "category": "编程套餐", "date": "2026-09", "lead": "Scott Wu（Cognition 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "被 Cognition 收购后更名：AI 原生 IDE 加 Agent 指挥中心，本地与云端代理并存。", "findings": [ "定价：Pro ≈$20/月起；更高档与 Teams 按座位。", "兼容：Agent Client Protocol（ACP），可运行 Claude Code、OpenCode、Codex 等代理。", "特点：代码库导航与多代理管理强，适合大型项目。", "迁移提示：原 Windsurf 订阅与账号体系已并轨，续费前看清新条款。", "避坑：多代理并行同时消耗额度，团队用先约定配额。" ], "source": "https://cognition.ai", "link": null, "slot": null }, { "id": "X-090", "title": "JetBrains AI", "en": "JETBRAINS AI", "department": "套餐 · JetBrains", "category": "编程套餐", "date": "2026-09", "lead": "Kirill Skrygan（JetBrains CEO）", "clearance": "REFERENCE AREA", "abstract": "JetBrains 官方 AI 订阅：自家 IDE 内直连，多代理可切换（含 Codex、Claude 与自带的 Junie）。", "findings": [ "档位（2026 参考）：AI Pro ≈$10/月起；更高档按可用额度与代理数递增，以官网为准。", "形态：IDE 内联补全 + 对话 + 多代理编排，JetBrains 全家桶用户几乎没有额外配置成本。", "代理可切换：能把外部代理（Codex / Claude 等）接进 IDE，用自己的 key 或官方额度。", "与同类对比：它卖的是「IDE 生态内的一体化」，不是最便宜的额度。", "避坑：额度与代理数绑定，重度用户先算清每月够不够。" ], "source": "https://www.jetbrains.com/ai/", "link": null, "slot": null }, { "id": "X-091", "title": "Google Antigravity 订阅", "en": "ANTIGRAVITY PLANS", "department": "套餐 · 国际", "category": "编程套餐", "date": "2026-09", "lead": "Demis Hassabis（DeepMind 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "谷歌的多代理工具，随 Google AI Pro（≈$20/月）提供，含浏览器控制与定时任务。", "findings": [ "定价：随 Google AI Pro 订阅（≈$20/月）提供。", "涉及模型：Gemini 系列。", "特点：多代理编排、浏览器控制、定时任务强。", "与 Gemini CLI 的关系：免费档自 2026-06 起迁移至 Antigravity CLI。", "避坑：别按老教程硬装 Gemini CLI 免费档，直接装 Antigravity。" ], "source": "https://antigravity.google", "link": "#/docs/antigravity", "slot": null }, { "id": "X-092", "title": "Trae 套餐", "en": "TRAE PLANS", "department": "套餐 · 预算友好", "category": "编程套餐", "date": "2026-09", "lead": "梁汝波（字节跳动 CEO）", "clearance": "REFERENCE AREA", "abstract": "VS Code 分支的平价 AI IDE：Lite ≈$3/月起，另有免费额度。", "findings": [ "定价：Lite ≈$3/月起，另提供免费额度。", "特点：入门成本最低的 AI IDE 之一，界面与 VS Code 习惯一致。", "涉及模型：多家主流模型（含国产与国际）。", "适合：预算有限但想用 IDE 内代理的开发者。", "避坑：低价档额度小，重任务建议配合自带 Key 或国内 Coding Plan。" ], "source": "https://www.trae.ai", "link": "#/docs/trae", "slot": null }, { "id": "X-093", "title": "OpenHands", "en": "OPENHANDS", "department": "套餐 · 开源代理", "category": "编程套餐", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "自主任务代理平台（开源）：在沙箱里跑完整任务，提供 SDK，偏企业向与自动化流水线。", "findings": [ "形态：自托管服务 + 沙箱执行，任务粒度比「编辑器里的代理」更大。", "可接任意模型（含本地模型），企业侧看重它的隔离与审计能力。", "适合：批量代码任务、CI 里的自动化修 bug、需要审计执行过程的团队。", "与编辑器插件的差别：它不依附 IDE，跑在服务器上。", "避坑：沙箱与权限要自己配好，别让代理直接碰生产仓库。" ], "source": "https://github.com/All-Hands-AI/OpenHands", "link": null, "slot": null }, { "id": "X-094", "title": "火山引擎方舟 Coding Plan", "en": "VOLCENGINE ARK", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "梁汝波（字节跳动 CEO）", "clearance": "REFERENCE AREA", "abstract": "字节的多模型聚合套餐：豆包、DeepSeek、Kimi、GLM、MiniMax 都能调，兼容主流工具。", "findings": [ "定价：Lite ≈¥40/月（首月常 ¥9.9，约 1200 次 / 5 小时、1.8 万次 / 月）；Pro ≈¥200/月（首月常 ¥49.9，约 6000 次 / 5 小时、9 万次 / 月）。", "Agent Plan：档位类似，额外支持多模态与 Agent 燃料值（AFP）。", "涉及模型：豆包、DeepSeek、Kimi、GLM、MiniMax 等。", "兼容：Claude Code、Cursor、OpenCode、Cline 等主流工具。", "避坑：热门档位常需抢购，条款对 Agent 滥用有限制。" ], "source": "https://www.volcengine.com/product/ark", "link": null, "slot": null }, { "id": "X-095", "title": "智谱 GLM Coding Plan", "en": "ZHIPU GLM PLAN", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "张鹏（智谱 CEO）", "clearance": "REFERENCE AREA", "abstract": "以 GLM 为主力的编码套餐，MCP 工具支持多，兼容 Claude Code 接入方式。", "findings": [ "定价：Lite ≈¥49–118/月；Pro / Max ≈¥149–500+（额度递增）。", "涉及模型：GLM 系列（含编码专用档）。", "兼容：Claude Code、CC Switch、Cursor 等；Anthropic 兼容端点用官方专用路径。", "适合：编码能力强、工具适配广，国产替代的常见首选。", "避坑：兼容端点路径随手改会 404，照官方文档填。" ], "source": "https://bigmodel.cn", "link": "#/docs/glm-coding", "slot": null }, { "id": "X-096", "title": "MiniMax Token Plan", "en": "MINIMAX PLAN", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "闫俊杰（MiniMax 创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "按 token 计费的多模态套餐：入门档 ≈¥29 起，TPS 高。", "findings": [ "定价：Starter / 入门 ≈¥29–49/月（数亿 token）；更高档 ≈¥119–469/月。", "涉及模型：MiniMax 系列（含多模态与 H 系列）。", "兼容：OpenCode、Cursor 等主流工具。", "特点：TPS 高、性价比好。", "避坑：token 制要盯住长上下文任务的消耗速度。" ], "source": "https://www.minimax.io", "link": "#/docs/minimax", "slot": null }, { "id": "X-097", "title": "Kimi 编程套餐", "en": "KIMI PLAN", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "杨植麟（月之暗面创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "月之暗面的编程订阅：≈¥49 起，长上下文是卖点。", "findings": [ "定价：≈¥49 起，不同档位对应请求 / 周限额，含会员权益。", "涉及模型：K 系列（长上下文见长）。", "兼容：官方 CLI（单文件版）与主流兼容工具。", "适合：长文档、长上下文与多步代理任务。", "避坑：认准 code.kimi.com 官方脚本，npm 上的同名包是无关抢注包。" ], "source": "https://code.kimi.com", "link": "#/docs/kimi-cli", "slot": null }, { "id": "X-098", "title": "阿里云百炼 Token / Coding Plan", "en": "ALIBABA BAILIAN", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "吴泳铭（阿里集团 CEO）", "clearance": "REFERENCE AREA", "abstract": "以通义千问为主并聚合其他模型，与阿里云生态集成好。", "findings": [ "定价：Lite 等档 ≈¥40/月（首月常 ¥7.9–39），按积分或请求限额；更高档与团队版按座位或更大 Credits。", "涉及模型：通义千问系列，并聚合其他模型。", "兼容：Claude Code、Cursor、OpenCode 等。", "适合：已用阿里云 / 钉钉体系的团队。", "避坑：积分制要看清单次任务的积分消耗。" ], "source": "https://bailian.console.aliyun.com", "link": "#/docs/qwen-code", "slot": null }, { "id": "X-099", "title": "腾讯云 Token / Coding Plan", "en": "TENCENT CLOUD PLAN", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "马化腾（腾讯创始人）", "clearance": "REFERENCE AREA", "abstract": "腾讯生态友好的低门槛套餐：混元等模型，≈¥28–40 起。", "findings": [ "定价：≈¥28–40 起，积分制或请求限额。", "涉及模型：混元 Hunyuan 等。", "兼容：CodeBuddy 三形态（IDE / 插件 / CLI）与主流工具。", "适合：已用腾讯云、企业微信生态的团队。", "避坑：国内版与国际版账号额度不通用。" ], "source": "https://cloud.tencent.com", "link": "#/docs/codebuddy", "slot": null }, { "id": "X-100", "title": "百度千帆 Token Plan", "en": "BAIDU QIANFAN PLAN", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "李彦宏（百度创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "档位最细的国产套餐：Mini / Lite / Pro / Max，≈¥4.9–200。", "findings": [ "定价：Mini / Lite / Pro / Max 四档，≈¥4.9–200，积分制。", "涉及模型：文心 ERNIE 系列（部分小模型已开源）。", "兼容：适配 Claude Code 等主流工具。", "适合：想用极低成本试水编程套餐的人。", "避坑：最低档额度有限，别指望跑长任务。" ], "source": "https://qianfan.cloud.baidu.com", "link": null, "slot": null }, { "id": "X-101", "title": "小米 MiMo Token Plan", "en": "XIAOMI MIMO PLAN", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "罗福莉（MiMo 负责人）", "clearance": "REFERENCE AREA", "abstract": "Credits 制、无严格 5 小时窗口，≈¥39 起。", "findings": [ "定价：≈¥39 起，Credits 制。", "特点：无明显 5 小时窗口限制，节奏更自由。", "涉及模型：MiMo 系列。", "兼容：Cursor 等主流工具。", "避坑：无窗口限制不等于不限量，仍要看月度 Credits 总额。" ], "source": "https://www.xiaomi.com", "link": null, "slot": null }, { "id": "X-102", "title": "其他国内平台合集", "en": "OTHER CN PLANS", "department": "套餐 · 国产", "category": "编程套餐", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "讯飞星辰、天翼云、七牛云、京东云、UCloud、华为 CodeArts、摩尔线程、无问芯穹、团结引擎、CSDN 星图等。", "findings": [ "科大讯飞星辰 Astron：≈¥19–200/月，积分制、多模型，首月常优惠。", "中国电信天翼云：≈¥40/月起，请求次数限额，支持 GLM 等模型。", "七牛云：企业向为主（个人相关资源包 ≈¥39）；京东云 / UCloud / 华为 CodeArts / 摩尔线程 / 无问芯穹 / 团结引擎约 ¥19.9–40 起。", "CSDN 星图等转售平台：常转售 GLM 等，价格更低但库存紧张、限制较多。", "避坑：首月优惠多（¥7.9–9.9 常见），但续费价与条款差异大，购买前逐条看限额与禁止项。" ], "source": "https://www.xfyun.cn", "link": null, "slot": null }, { "id": "X-103", "title": "DeepSeek vs 千问", "en": "DEEPSEEK vs QWEN", "department": "对比 · 开源旗舰", "category": "对比合集", "date": "2026-09", "lead": "梁文锋（DeepSeek CEO）、吴泳铭（阿里集团 CEO）", "clearance": "REFERENCE AREA", "abstract": "两条国产开源主线正面对照：V4 系列 vs Qwen3.8 系列，比编码与 agent、比许可、比本地可跑性。", "findings": [ "许可差异：DeepSeek 主力档多用 MIT（最宽松）；Qwen 中小尺寸为 Apache-2.0，旗舰 Qwen3.8-Max 的权重以自定义许可开放。", "形态差异：DeepSeek 偏「少而精」的旗舰路线（V4-Flash / V4-Pro / V4.1-Flash）；Qwen 是「全尺寸铺开」（27B 稠密到 2.4T-A95B 稀疏）。", "本地可跑性：Qwen 的 27B 档在消费级显卡上更现实；DeepSeek 满血档普遍要多卡或大内存工作站。", "定价风格：DeepSeek 因峰谷定价波动被社区反复调侃；Qwen 走 API + 云厂商捆绑，价格随平台活动变动。", "结论取向：要「能塞进自己机器 + 宽松许可」看 Qwen 中小档；要「少买卡、直接调最强开源旗舰」看 DeepSeek V4 系列。" ], "source": "/docs/compare/deepseek-vs-qwen", "link": "#/docs/compare/deepseek-vs-qwen", "slot": null }, { "id": "X-104", "title": "Cursor vs Windsurf", "en": "CURSOR vs WINDSURF", "department": "对比 · AI IDE", "category": "对比合集", "date": "2026-09", "lead": "Michael Truell（Cursor CEO）、Scott Wu（Cognition CEO）", "clearance": "REFERENCE AREA", "abstract": "两款 AI 原生 IDE 的定位与价格对照：一个走「编辑器 + 代理深度整合」，一个主打「代理优先的工作流」。", "findings": [ "产品定位：Cursor 以编辑器体验为核心增量（补全、改写、Agent 模式并存）；Windsurf 更强调代理自动跑完整任务。", "价格结构：两家都是免费档 + 个人订阅 + 团队档，档位与额度随版本调整，购买前看官网当前定价页。", "国内可用性：都依赖其云端服务；网络不稳时体验下降明显，接入本地模型的能力有限。", "替代路线：预算敏感或要求数据不出本机，可换 VS Code 系插件（Cline 一类）+ 本地模型，见站内对应文档。", "结论取向：重度写代码、要「手感」选 Cursor；想把整段任务交给代理跑，试 Windsurf 的代理流。" ], "source": "/docs/compare/cursor-vs-windsurf", "link": "#/docs/compare/cursor-vs-windsurf", "slot": null }, { "id": "X-105", "title": "Claude Code vs Codex CLI", "en": "CLAUDE CODE vs CODEX CLI", "department": "对比 · 终端代理", "category": "对比合集", "date": "2026-09", "lead": "Dario Amodei（Anthropic CEO）、Sam Altman（OpenAI CEO）", "clearance": "REFERENCE AREA", "abstract": "两个官方终端代理的对照：能力侧重、额度与国内可安装性。", "findings": [ "安装与网络：两者都能用 npm 安装并走国内镜像；官方脚本在受限网络下不可直连（详见各自文档）。", "能力侧重：Claude Code 以长任务与工具编排见长；Codex CLI 与 ChatGPT 订阅绑定更紧，轻量任务更省。", "计费方式：分别绑定 Anthropic / OpenAI 的订阅或 API 额度，额度耗尽后的行为不同。", "混用建议：两条都装并不冲突——重活走一个、日常小改走另一个，站内一键应用可把两条都配好。", "避坑：别把「能装」当「能用」——网络与额度才是日常体验的分水岭。" ], "source": "/docs/compare/claude-code-vs-codex", "link": "#/docs/compare/claude-code-vs-codex", "slot": null }, { "id": "X-106", "title": "Ollama vs LM Studio", "en": "OLLAMA vs LM STUDIO", "department": "对比 · 运行器", "category": "对比合集", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "本地推理两条主流路线：命令行派（Ollama）与图形界面派（LM Studio）怎么选。", "findings": [ "Ollama：一条命令拉起 + OpenAI 兼容 API，适合接进脚本与现有工具链。", "LM Studio：图形界面友好，内置模型搜索与下载，适合先跑通再谈优化。", "性能并不由界面决定：同一模型同一量化档，两者差距通常小于量化档位本身带来的差距。", "进阶路线：要极致控制用 llama.cpp，要多用户生产环境用 vLLM（见站内文档）。", "结论取向：先 LM Studio 跑通，再按需切 Ollama 接自动化。" ], "source": "/docs/kb/ollama-vs-lmstudio", "link": "#/docs/kb/ollama-vs-lmstudio", "slot": null }, { "id": "X-107", "title": "本地跑 vs 走 API", "en": "LOCAL vs API", "department": "对比 · 部署路线", "category": "对比合集", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "成本、隐私、门槛三条线对照：什么时候该在自己机器上跑，什么时候该直接调 API。", "findings": [ "成本结构：本地是一次性买卡 + 电费，API 是按量付费；跑得少大概率 API 更便宜，跑得多且稳定才轮到本地。", "隐私与合规：本地档数据不出本机；API 一定上传云端（条款里对训练与留存的规定各家不同，逐个看）。", "能力上限：闭源旗舰（GPT-6 / Claude / Gemini 一线）仍普遍强于可本地跑的开放权重档，硬推理类任务差距更明显。", "可用性风险：API 受网络与配额影响，本地受显存与维护成本影响——两者都会「卡」，只是卡的地方不同。", "结论取向：研究与隐私场景优先本地；产品化与峰值能力优先 API；混合（本地跑日常、API 打高峰）是最常见的实际做法。" ], "source": "/docs/compare/local-vs-api", "link": "#/docs/compare/local-vs-api", "slot": null }, { "id": "X-108", "title": "Windsurf", "en": "WINDSURF", "department": "IDE · AI 原生", "category": "工具箱·终端", "date": "已上线", "lead": "Scott Wu（Cognition 联合创始人兼 CEO）", "clearance": "REFERENCE AREA", "abstract": "代理优先的 AI 原生 IDE：把整段任务交给代理跑，同时保留编辑器形态。", "findings": [ "定位与 Cursor 的差别：它更强调「代理自动跑完整任务」，Cursor 更偏编辑器体验。", "同构定价（免费档 + 个人 + 团队），额度用尽后的降级行为差异大，购买前看官网定价页。", "依赖云端服务，网络不稳时体验明显下降；接入本地模型的能力有限。", "想先比一轮再决定，看站内对比《Cursor vs Windsurf》。" ], "source": "/docs/windsurf", "link": "#/docs/windsurf", "slot": null }, { "id": "X-109", "title": "Trae Agent", "en": "TRAE AGENT", "department": "开源代理 · 字节", "category": "工具箱·国产", "date": "已上线", "lead": "梁汝波（字节跳动 CEO）", "clearance": "REFERENCE AREA", "abstract": "字节开源（MIT）的 SWE 代理：命令行跑任务，BYOK 自带模型 key，适合绕开 IDE 的资源占用。", "findings": [ "开源可自部署：不依赖 Trae IDE 客户端，介意遥测与资源占用的人可以只用这一条。", "BYOK 模式：模型 key 用自己的（可接国内模型或本地模型），成本与数据边界都自己控。", "能力面向「改代码 + 跑命令 + 迭代到通过」的 SWE 任务，不是通用聊天。", "与 Cline / OpenCode 一类同赛道，差异在官方维护与工具链完整度。" ], "source": "/docs/trae-agent", "link": "#/docs/trae-agent", "slot": null }, { "id": "X-110", "title": "通义灵码（历史档）", "en": "LINGMA", "department": "插件 · 阿里（已并入 Qoder）", "category": "工具箱·国产", "date": "已上线", "lead": "吴泳铭（阿里集团 CEO）", "clearance": "REFERENCE AREA", "abstract": "上一代阿里编码助手：以 IDE 插件形态为主，2026 年品牌与产品线整合进 Qoder。", "findings": [ "现役产品是 Qoder（站内另有单档）；本档保留给仍在用旧插件的用户查迁移路径。", "旧插件的免费策略与额度已随整合调整，新用户建议直接看 Qoder。", "企业侧原走阿里云体系（百炼），迁移口径以官方公告为准。" ], "source": "/docs/tongyi-lingma", "link": "#/docs/tongyi-lingma", "slot": null }, { "id": "X-111", "title": "Zed", "en": "ZED", "department": "套餐 · 编辑器", "category": "编程套餐", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "Rust 原生编辑器：启动与输入延迟极低，Pro 档含托管 token，可用 ACP 接外部代理。", "findings": [ "档位（2026 参考）：Personal 免费；Pro ≈$10/月（含部分托管 token），以官网为准。", "形态：编辑器本体轻快，AI 能力走「自带 token + 接外部代理」两条路。", "ACP（Agent Client Protocol）：能把 Claude Code、Codex 一类代理接进编辑器，属于它最实用的卖点。", "与 JetBrains AI 的差别：它更偏「轻编辑器 + 开放接代理」，不是全家桶一体化。", "避坑：托管 token 用尽后按量计费，重度使用先估月消耗。" ], "source": "https://zed.dev", "link": null, "slot": null }, { "id": "X-112", "title": "Amazon Kiro", "en": "KIRO", "department": "套餐 · 编辑器", "category": "编程套餐", "date": "2026-09", "lead": "Andy Jassy（Amazon CEO）", "clearance": "REFERENCE AREA", "abstract": "主打 spec-driven（先写规格再开工）的 AI IDE：把「需求 → 设计 → 任务」做成可审阅的文档再动代码。", "findings": [ "档位（2026 参考）：Pro ≈$20/月起，含一定额度，超出按量，以官网为准。", "方法论差异：它强制你先落规格与任务清单，适合「需求要留痕」的团队。", "对比同类：Cursor / Windsurf 更偏「边聊边改」，Kiro 偏「先立文档再执行」。", "托管在 AWS 体系里，企业侧账号与计费好归口。", "避坑：spec 流程前期更慢，赶时间的小改动会觉得重。" ], "source": "https://kiro.dev", "link": null, "slot": null }, { "id": "X-113", "title": "Continue", "en": "CONTINUE", "department": "套餐 · 开源代理", "category": "编程套餐", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "开源 IDE 助手（插件形态）：本地优先，可直接接 Ollama 一类本地模型，也能接云端 API。", "findings": [ "开源可自建：VS Code / JetBrains 插件，配置与模型全由自己掌控。", "本地优先：默认就能接本地模型，数据不出本机这一条对隐私敏感场景很实用。", "与 Cline 的差别：Continue 更偏「助手」（补全 / 对话 / 编辑），Cline 更偏「自主代理」。", "配合站内一键应用可把本地模型与它一起配好。", "避坑：插件形态能力受宿主 IDE 限制，长任务不如独立代理稳。" ], "source": "https://github.com/continuedev/continue", "link": null, "slot": null }, { "id": "X-114", "title": "Goose", "en": "GOOSE", "department": "套餐 · 开源代理", "category": "编程套餐", "date": "2026-09", "lead": "—", "clearance": "REFERENCE AREA", "abstract": "MCP 原生的开源代理：桌面应用 + CLI 双形态，擅长把外部工具接成复杂工作流。", "findings": [ "MCP 原生：工具接入是它的核心设计，适合「一个任务要调很多外部服务」的场景。", "形态：桌面应用上手快，CLI 便于自动化；可配本地或云端模型。", "与 OpenHands 的差别：OpenHands 偏服务器侧沙箱执行，Goose 偏本机工作流编排。", "开源可自建，企业可内网部署。", "避坑：工具权限要逐项确认，别一次授予全盘读写。" ], "source": "https://github.com/block/goose", "link": null, "slot": null } ] }