啪叽

@melvynx 🙋

who still use opencode ?

@coder_left lite

@lbf911 lite还是pro，pro也这样那还用个蛋😅

尼玛，方舟 Coding Plan 价格打到9块9了 官方宣称，不仅支持主流大语言模型，还支持 Embedding 向量化模型，涵盖 Doubao Seed Code、Doubao Seed 2.0 系列、GLM 系列、MiniMax 系列、Kimi系列、DeepSeek 系列、Doubao Embedding Vision 等模型 lite 套餐原价40元，现在只用9块9 pro 套餐原价200元，现在只用49块9 有没有用过的兄弟聊聊体验如何？🤔

@KimiDevs 所以最适合kimi-code的模型是哪个呢？

Hello world!

@real_kai42 前天让agy介绍一下kimi-code，它给介绍了codex哈哈哈

啊？◔‸◔?

@Young_AGI 体验了一下，感觉还挺好用的，还不错，哈哈哈。要是能换theme就好了

Kimi Code Changelog --- 0.11.0 🚀 （Less than 20 hours since the last version 0.10.0 update.）🫡

@OpenCodeLog 有个security_的py的hook一直执行

𝙊𝙥𝙚𝙣𝘾𝙤𝙙𝙚 v1.16.0 released. TL;DR: v2 sessions move into core, App/TUI get stronger workspace controls, providers get protocol fixes, and Stats gets model-level pages. 𝗖𝗼𝗿𝗲 • Added a public native core API for Effect-based embeddings: sessions can now be created, listed, prompted, paged, inspected, and streamed through an intentional supported surface. • Changed v2 session execution around event-sourced prompt admission, queued vs steering delivery, durable tool settlement, and location-scoped model/tool resolution. 𝗧𝗨𝗜 Added • Added /move for moving a session to another project directory or creating a new git worktree copy, with dirty-file transfer handling. • Added queued prompt management in direct interactive run mode, including leader+q access plus edit/remove before queued prompts start. • Added Vue syntax highlighting. Fixed • Fixed question replies and diff views so they route through the active session directory instead of the current process directory. • Fixed prompt corruption when pasting near wide characters. • Fixed live hydration and session switching edge cases that could leave stale or misplaced active-turn UI. 𝗔𝗽𝗽 Added • Added Settings v2 appearance controls for color scheme, theme selection, UI font, mono font, and terminal font. • Added server settings and multi-server home project lists, including local/http server de-duping and per-server project state. • Added home session search with keyboard navigation, status indicators, and project-aware results. Fixed • Fixed recent session/project loading before path sync resolves, stale custom provider state after config updates, and titlebar/tab layout polish. 𝗗𝗲𝘀𝗸𝘁𝗼𝗽 • Fixed local server startup failure handling so sidecar initialization errors surface in the renderer instead of hanging behind startup state. • Changed startup by removing the old SQLite migration overlay path and waiting directly on sidecar readiness. 𝗦𝗲𝗿𝘃𝗲𝗿 • Added a standalone v2 server package with typed routes for sessions, messages, models, providers, filesystem, permissions, questions, events, commands, skills, and health. • Added experimental project-copy and control-plane routes for managed working copies and session moves. 𝗟𝗟𝗠 Added • Added chronological system-message support: Anthropic Opus 4.8 can use native system updates, while other routes receive visible lower-authority wrapped updates. Fixed • Fixed OpenAI function tool schemas by normalizing top-level objects and flattening nullable/anyOf shapes. • Fixed Gemini and Bedrock reasoning signature preservation across replay and continuation. 𝗣𝗿𝗼𝘃𝗶𝗱𝗲𝗿𝘀 Added • Added AWS Bedrock Mantle support for OpenAI models, including OpenAI Responses routing for GPT models and Chat routing for safeguard models. Changed • Updated GitHub Copilot model discovery to use the 2026-06-01 API, picker filtering, token-pricing conversion, and utility small-model selection. Fixed • Fixed SAP AI Core reasoning variants by routing Anthropic, Gemini, GPT, and fallback efforts through modelParams. • Fixed OpenAI websocket error handling so initial and mid-stream API errors preserve status, headers, and response bodies. 𝗔𝗖𝗣 • Fixed loaded-session replay so ACP clients receive prior user, assistant, and reasoning transcript chunks. • Fixed session/cancel so ACP aborts the running OpenCode turn while keeping the ACP session usable. • Fixed ACP tool classification and locations for apply_patch, task, external_directory, and cleaned read tool display content. 𝗦𝘁𝗼𝗿𝗮𝗴𝗲 • Changed database ownership into core with a named TypeScript migration journal and normalized Windows path storage. • Removed the manual JSON-to-SQLite migration command from opencode db. 𝗖𝗼𝗻𝘀𝗼𝗹𝗲 • Added refreshed Stats routes with cache-ratio, geo breakdown, top-usage sorting, and model/provider detail pages. • Fixed Stats banner proxying, OG image serving, and retired provider/model cleanup during sync. 𝗭𝗲𝗻 • Added Go/Zen catalog coverage for MiniMax M3, Qwen3.7 Plus, Qwen3.7 Max, and DeepSeek V4 Flash surfaces. • Updated the free NVIDIA endpoint from Nemotron 3 Super to Nemotron 3 Ultra with revised trial/TOS copy. 𝗣𝗹𝘂𝗴𝗶𝗻 Added • Added experimental.provider.small_model so provider plugins can choose utility models for tasks like title generation. Changed • Changed plugin event delivery to location-scoped EventV2 events instead of the legacy global bus. Fixed • Restored private git plugin install fallback behavior. 𝗨𝗜 • Added OC-2 v2 theme token resolution, Matrix theme, and a shared ProjectAvatar v2 component. • Improved v2 select, menu, toast, tabs, keybind, switch, and text-input styling for the new App settings/home surfaces. 𝗦𝗗𝗞 • Regenerated OpenAPI and the JS v2 SDK for the new v2 server routes and changed response shapes. 𝗭𝗲𝗱 • Removed the bundled Zed extension metadata and sync automation from this repository. No noticeable bundle change Compare: github.com/anomalyco/open…

@lidangzzz 有类似omo的插件吗

主观体验来看，kimi code比opencode好用

@real_kai42 @mranti 必须得用kimi的模型吗？

欢迎大家关注下我们新的开源项目 kimi-code 链接在评论区了 （🥲 开源完就病倒了，今天精神好多了，好像错过了 x 上的热闹时候）

233333

@BruceBlue 非常慢，咋能推荐的下去😂

火山方舟的 Coding Plan 这么便宜？买一个 Lite 实测一下 Kimi K2.6🤔 智谱 GLM Lite ¥44 才 80 次/5小时，MiniMax Starter 要$10，Kimi 自家最低档也得 $15 起步…… 而火山 Lite 只要¥40（不到$6），每5小时约 1200 次请求！Pro ¥200（不到$30）5倍用量。 我主要用 Kimi K2.6 在 Hermes Agent 里测试，感受如下。 🔹国内优秀模型不用抢：套餐内包含 Kimi-K2.6、GLM-5.1、MiniMax-M2.7 等比较有优秀的模型； 🔹工具全兼容：Cursor、Claude Code、OpenClaw、Hermes Agent等，额度共享； 🔹用量消耗：昨天订阅后，轻度使用（2个Agent系统Code Review + Bug修复）花了10%额度。 如果狠狠用几天，一周不到大概率用完；重度用户建议直接上 Pro； 🔸缺点：偶尔会因为“用的人太多”触发 429 限流（尤其是Kimi-K2.6和GLM-5.1），Auto 模式会智能避开高消耗模型，可能会调用字节的豆包模型🐶。 ✅老实说：Kimi K2.6 在方舟上表现极强，Bug 定位/修复速度和解决问题的能力保底优于 Claude Sonnet 4.6，达到了 Claude Opus的～90%左右。 👉新用户首单有 9 折：volcengine.com/L/a4d2ATBQME4 #kimi #GLM #MiniMax #HermesAgent

@mranti 没有，巨慢

不是，Nvidia免费提供的LLM，访问速度怎么这么慢啊？我从来没顺畅访问过。有人访问的顺畅吗？

@LocationCk @Pluvio9yte 我也是

@Pluvio9yte ghostty用完基本回不去iterm了啊

ghostty和iterm2，大家更推荐哪一个呢

@shenjiangping @cnfinancewatch 好像签了无固定期限合同之后，貌似协商一致也是n+1

@cnfinancewatch 深圳这边我看到的非大厂的情况，二签三签没区别，身边好些程序员朋友签了三签，裁员的时候照样只有n+1。去年至今认识的被裁的都接受了n+1走人

大厂的潜规则： 一般情况，二签到期后，不是不可替代的人就不会三签了，人力成本的问题，知识和技术更新的问题。企业也是无奈，签了无固定期后，更容易躺平。当然真正能不断进步的员工，公司是舍不得裁员的。

@laogui @skywind3000 降智特别明显

@skywind3000 用了火山和阿里的 coding plan，表示以后再也不买这些玩意了，除了他们自家地模型，第三方模型都降智，还是要买模型厂商官方的套餐才行。

我抢了一周腾讯云和 glm 的 coding plan，每天早上十点正准时开抢，就是抢不到，十多秒就没了，好几天操作到一半就失败了，不知道有没有平替？

@lo114_ 细说

kimi面试体验太好了 就是我太菜了😭

@chaoren23333 @geekbb 我昨天刚买了一个母号，送5个子号

@geekbb 那我充值的两个team算什么😭

喜报，今天 OpenAI 免费帐号又可以蹬 GPT-5.4 啦🥳

@huangyun_122 我用这个总是断流，然后在cli上就没有，怀疑是不是我vps的问题

不行了兄弟们，这个开源库必须分享 —— Sub2API 它的神奇功能，只能用 breaking 来形容： 1/ 把 200刀的 Claude 订阅分享给你所有的好友用，简称“拼车”。其他大模型， GPT, Gemini 也都可以 2/ 不舍得用 200刀的朋友，可以买 2个 20刀的，嘿嘿，大部分人不知道 3/ 胆子大点，或许你可以，嘿嘿…… 链接放评论区

@m1ssuo 容易封ip😂

暂时性 Token 自由了。

@fkysly codex感觉给我把ip限制了，在本地cliproxyapi上可以，在服务器上的sub2api里就503

我不是把所有账号都搞了一个阿里云轻量服务器，然后走 sub2api 了，等于自己搭建了一个内部中转站么？还带了几个哥们一起用。 已经将近1个星期了，还挺稳定，我光 anthropic 账号就挂了4个在上面，感觉还不错啊。虽然 sub2api 比较挫，saas 功能一团糟，还有不知名限流的 bug，不过总体来说，这套体验还是不错的。不需要管这账号那账号了，没人有个统一的 api-key 通过 cc-switch 管理就能用。