第 024 期
2026 年 10 月 5 日 · …
28 天冲刺:28 天内每天一个有感改进,否则全员 reset;口碑倒向 Opus 5.5。Grok 4.7 双榜第一:在 Frontier v4 与 Cyber Index 双榜第一。pstack 推 /correct:DHH 让 agent 把 Campfire 移植到 5 种技术栈。
28 天 Codex 每日改进冲刺
CodexOpus 5.5Grok 4.7pstackDHH
Issue 024
Oct 05, 2026 · …
28-day sprint: One meaningful daily improvement or full team reset for 28 days; sentiment tilts to Opus 5.5.Grok 4.7 tops two charts: First on both Frontier v4 and Cyber Index leaderboards.pstack ships /correct: DHH has agents port Campfire to 5 tech stacks.
28 days Codex daily improvement sprint
CodexOpus 5.5Grok 4.7pstackDHH
第 023 期
2026 年 10 月 4 日 · …
Codex 征集痛点:Tibo 公开问「Codex 还缺什么」,OpenCode 2 同步征集槽点。OpenClaw Android 卡审超一周:Pi Durable 在手机上跑多人 agent。模型进 IDE:Antigravity 上 Opus 5.5 / Sonnet 5.5;Cline 免费接入 Ling 3.1 Flash。
1 周+ OpenClaw Android 卡在审核
CodexOpenCodeOpenClawClinePi Durable
Issue 023
Oct 04, 2026 · …
Codex crowdsources: Tibo publicly asks 'what's Codex missing'; OpenCode 2 runs a parallel pain-point survey.OpenClaw Android stuck in review: Over a week in Google review; Pi Durable runs multi-user agents on phones.Models enter IDEs: Antigravity adds Opus 5.5 / Sonnet 5.5; Cline ships free Ling 3.1 Flash.
1 wk+ OpenClaw Android stuck in review
CodexOpenCodeOpenClawClinePi Durable
第 022 期
2026 年 10 月 3 日 · …
GPT-6.1 Sol 恢复:付费 ChatGPT 全局 reset,速度回到预期。Claude Mods 可演示:DeepSeek 推桌面版 Harness,Pi Durable 接入 Cloudflare。同权重换 harness:62% vs 33%;T3 Code 到 40 万用户;Muse Gadgets 开源。
62 : 33 同一权重,不同 harness 的得分
GPT-6.1 SolClaude ModsT3 CodeDeepSeekPi Durable
Issue 022
Oct 03, 2026 · …
GPT-6.1 Sol recovers: Paid ChatGPT global reset; speed returns to expected levels.Claude Mods demo-ready: DeepSeek ships desktop Harness; Pi Durable integrates Cloudflare Agents SDK.Same weights, different harness: 62% vs 33% on the same model; T3 Code hits 400K users; Muse Gadgets open-sourced.
62 : 33 Same weights, different harness scores
GPT-6.1 SolClaude ModsT3 CodeDeepSeekPi Durable
第 021 期
2026 年 10 月 2 日 · …
Claude Mods 上线:用 TypeScript 改行为 / UI / 功能,插件可分享;design / deck / doc 享 50% 优惠。GPT-6.1 Sol 爆棚:需求爆棚、OpenAI 紧急加容量;浏览器任务分更高、缓存令牌更低。Pi 1.0 / Durable:Pi 1.0 与 Pi Durable 并行;Cursor 接入 GLM 5.3;Factory Automations GA。
50% Claude App design/deck/doc 用量优惠
Claude ModsGPT-6.1 SolPiCursorFactory
Issue 021
Oct 02, 2026 · …
Claude Mods launches: Modify behavior / UI / features with TypeScript; plugins are shareable; 50% off design / deck / doc.GPT-6.1 Sol overloaded: Demand surges; OpenAI emergency adds capacity; higher browser task scores, lower cached token costs.Pi 1.0 / Durable: Pi 1.0 and Pi Durable run in parallel; Cursor adds GLM 5.3; Factory Automations goes GA.
50% Claude App design/deck/doc usage discount
Claude ModsGPT-6.1 SolPiCursorFactory
第 020 期
2026 年 10 月 1 日 · …
Gemini 4 Argon 发布:Google / DeepMind 新 frontier,主打长程软件工程,宣称约 1M token 输出上限。Factory ↔ Cognition 冲突:Factory 解聘顾问并指控利益冲突;Cognition 同日宣布其任 CRO。Grok Bot 接 Cursor:可把编码任务交给 Cursor 并管理 PR;Dots 定价继续被官方澄清。
1M Gemini 4 Argon 输出 token 上限
Gemini 4FactoryCognitionGrok BotCursor
Issue 020
Oct 01, 2026 · …
Gemini 4 Argon launches: Google / DeepMind new frontier model, ~1M token output cap for long-range software engineering.Factory-Cognition clash: Factory fires advisor citing conflict of interest; Cognition announces him as CRO the same day.Grok Bot connects Cursor: Can hand coding tasks to Cursor and manage PRs; Dots pricing continues to be clarified.
1M Gemini 4 Argon output token cap
Gemini 4FactoryCognitionGrok BotCursor
第 019 期
2026 年 9 月 30 日 · …
DevDay 三连发:always-on 个人代理 Dots、GPT-6.1 Sol(约 1/5 Astra 价)、最高约 300 tok/s Ultrafast。订阅直通第三方:Sign in with ChatGPT 可把额度用到 Devin / OpenCode / Notion 等 16+ 产品。工具侧:Cursor 聊天内 /visualize;OpenClaw Enterprise 开源企业控制平面。
300 tok/s Ultrafast / Pro 500 最高速度
DotsGPT-6.1 SolCodexCursorOpenClaw
Issue 019
Sep 30, 2026 · …
DevDay triple launch: Always-on personal agent Dots, GPT-6.1 Sol (~1/5 Astra price), up to ~300 tok/s Ultrafast.Subscriptions reach third parties: Sign in with ChatGPT lets you use quota with 16+ products like Devin, OpenCode, Notion.Tools update: Cursor ships in-chat /visualize; OpenClaw Enterprise open-sources its control plane.
300 tok/s Ultrafast / Pro 500 top speed
DotsGPT-6.1 SolCodexCursorOpenClaw
第 018 期
2026 年 9 月 29 日 · …
Claude Sonnet 5.5 发布:相对 Sonnet 5 约快 30%、多数工作最多便宜约 30%;成为 Claude Code 默认。Theo 拆配额:约 $200 订阅≈$9k/月 Opus 用量;Opus 5.5 把 Astra 的 ts-rust 推倒重写。Devin 降价:Devin 降价并推 Mobile;OpenCode 谈 Go 经济性与免费 SSO。
$9k/月 Theo 拆算的 $200 Claude 订阅 Opus 用量
Sonnet 5.5Claude CodeDevinOpenCodeCursor
Issue 018
Sep 29, 2026 · …
Claude Sonnet 5.5 launches: ~30% faster, up to ~30% cheaper than Sonnet 5; becomes Claude Code default.Theo pegs quota value: ~$200 subscription ≈ $9k/mo of Opus usage; Opus 5.5 rewrites Astra's ts-rust from scratch.Devin cuts prices: Devin drops prices and ships Mobile; OpenCode talks Go economics and free SSO.
$9k/mo Theo's estimate of $200 Claude plan Opus usage
Sonnet 5.5Claude CodeDevinOpenCodeCursor
第 017 期
2026 年 9 月 28 日 · …
Opus 5.5 刷爆配额:从「魔法」体感到刷爆 Max 配额;相对 Fable 5.1 约 4–6× 限额体感。Cline 上架 Ember-1:基于 Kimi K3,约少 40% tokens;Copilot 强调并行 agents。UFO 多智能体 OS:多智能体 harness OS 宣布上线;另有 Jev-as-a-Judge。
4–6× Opus 5.5 相对 Fable 5.1 的限额消耗
Opus 5.5ClineCopilotUFOJev
Issue 017
Sep 28, 2026 · …
Opus 5.5 maxes quotas: From 'magic' impressions to burning through Max quotas; ~4-6x limit usage vs Fable 5.1.Cline ships Ember-1: Based on Kimi K3 with ~40% fewer tokens; Copilot highlights parallel agents.UFO multi-agent OS: Multi-agent harness OS launches; plus Jev-as-a-Judge.
4–6× Opus 5.5 quota usage vs Fable 5.1
Opus 5.5ClineCopilotUFOJev
第 016 期
2026 年 9 月 27 日 · …
Codex 宕机后重置:短暂宕机后恢复,并对付费用户重置限额。Claude 人气约两倍:T3 Code 用量显示 Claude 人气已是 Codex 的约两倍。Jev 硬刚 Astra:Jev Router 被拿来硬刚 GPT-6 Astra;OpenCode 强调「赢的是 harness」。
2× T3 Code 中 Claude 对比 Codex 的人气
CodexT3 CodeClaudeJevOpenCode
Issue 016
Sep 27, 2026 · …
Codex resets after outage: Brief outage followed by usage limit reset for all paid users.Claude ~2x more popular: T3 Code usage data shows Claude roughly twice as popular as Codex.Jev takes on Astra: Jev Router pitched against GPT-6 Astra; OpenCode insists 'the harness wins.'
2× Claude vs Codex popularity in T3 Code
CodexT3 CodeClaudeJevOpenCode
第 015 期
2026 年 9 月 26 日 · …
Copilot 史上最大更新:Autopilot / Code / Home / Office 并与 OpenClaw 合作落地企业部署。Devin 年化收入过 $1B:Cognition 称 Devin 年化经常性收入跑过 10 亿美元。Claude Code 新技巧:Slack 里的 Tag、5 小时限额优雅收尾、/checkup prompt-audit。
$1B Devin 年化经常性收入
CopilotDevinClaude CodeOpenClawOpus 5.5
Issue 015
Sep 26, 2026 · …
Copilot biggest update ever: Autopilot / Code / Home / Office plus enterprise deployment with OpenClaw.Devin ARR passes $1B: Cognition says Devin's annualized recurring revenue has passed $1 billion.Claude Code triple feature: Tag in Slack, graceful 5-hour limit handling, /checkup prompt-audit.
$1B Devin annualized recurring revenue
CopilotDevinClaude CodeOpenClawOpus 5.5
第 014 期
2026 年 9 月 25 日 · …
OpenClaw 砍 40 万行:用 test-audit skill 砍掉约 40 万行自测代码,覆盖率几乎不动。订阅体感翻转:Claude Code 与 Codex 的订阅体感再次翻转;Muse 上 Mac computer use。System One 扩展:从 Jev 扩到 CLM;Browser Use 用 Luna→Jev 打牌。
40 万行 OpenClaw 砍掉的自测代码
OpenClawClaude CodeCodexMuseOpenCode
Issue 014
Sep 25, 2026 · …
OpenClaw cuts 400K lines: test-audit skill removes ~400K lines of self-test code with almost no coverage loss.Subscription sentiment flips: Claude Code vs Codex sentiment flips again; Muse ships Mac computer use.System One expands: Grows from Jev to CLM; Browser Use plays the Luna→Jev routing card.
400K lines Self-test code OpenClaw removed
OpenClawClaude CodeCodexMuseOpenCode
第 013 期
2026 年 9 月 24 日 · …
Cursor 成本再降:agent token 成本再降约 7%,并公开一长段「改造 harness」提示词。Browser Use Bench 重画:GPT-6 Sol / Luna 与 Opus 5.5 重画帕累托前沿。harness 讨论升温:rauchg 的 Brain/Hands/Files、Google 自改进 harness 论文,Jev 用例拆解。
7% Cursor agent token 成本降幅
CursorGPT-6Opus 5.5JevBrowser Use
Issue 013
Sep 24, 2026 · …
Cursor costs drop again: Agent token costs cut by ~7%, plus a long 'harness customization' prompt shared publicly.Browser Use Bench redrawn: GPT-6 Sol / Luna and Opus 5.5 redraw the Pareto frontier.Harness debate heats up: rauchg's Brain/Hands/Files, Google's self-improving harness paper, Jev use case deep dive.
7% Cursor agent token cost reduction
CursorGPT-6Opus 5.5JevBrowser Use
第 012 期
2026 年 9 月 23 日 · …
Claude Opus 5.5 发布:对标 Fable 5.1 智力、比 Opus 5 更便宜;成为 Claude Code 默认模型。GPT-6 Sol / Luna 跟进:约一百分钟后发布,强调 coding / computer use,API 价格再砍约一半。OpenCode 2.0 亮相:谈「会话中实时改自身」,Grok 4.7 进入次日实战与批评。
~100 分钟 Opus 5.5 与 GPT-6 发布间隔
Opus 5.5GPT-6OpenCodeGrok 4.7Cursor
Issue 012
Sep 23, 2026 · …
Claude Opus 5.5 launches: Matches Fable 5.1 intelligence at lower cost than Opus 5; becomes Claude Code's default.GPT-6 Sol / Luna follows: Ships ~100 minutes later; emphasizes coding / computer use with ~50% API price cut.OpenCode 2.0 debuts: Talks about 'modifying itself mid-session'; Grok 4.7 enters next-day field testing and criticism.
~100 min Gap between Opus 5.5 and GPT-6 launches
Opus 5.5GPT-6OpenCodeGrok 4.7Cursor
第 011 期
2026 年 9 月 22 日 · …
Grok 4.7 发布:同价同速下明显强于 4.6;在 Cursor / Cline / Devin / Droid / VS 上线。Devin Cloud in Terminal:推出 devin ssh;同日把 Grok 4.7 接进 Cloud / Desktop / CLI。Cursor 一月合入约 2500 PR:Compile 缺席演讲变长文,poteto 复盘一个月的合入量。
2500 Cursor 一月合入的 PR 数
Grok 4.7DevinCursorClineMiMo
Issue 011
Sep 22, 2026 · …
Grok 4.7 launches: Clearly stronger than 4.6 at the same price and speed; live on Cursor / Cline / Devin / Droid / VS.Devin Cloud in Terminal: Ships devin ssh; same day adds Grok 4.7 to Cloud / Desktop / CLI.Cursor merges ~2500 PRs in a month: Compile talk becomes a blog post; poteto recaps the month's merge volume.
2500 PRs Cursor merged in one month
Grok 4.7DevinCursorClineMiMo
第 010 期
2026 年 9 月 21 日 · …
Jev 全面开放:取消 waitlist;同日开源 Kev 小模型家族与 DocJev 文档分类库,生态一天三连。Claude Code 省 token 工作坊:官方工作坊被日语圈疯狂转发;Devin SWE-2 云端限时免费。Harness Tax 论文:UC Berkeley 团队发现换 harness 成功率差不多,费用最多差 5 倍——Pi 最省。
5× 不同 harness 间的成本差异(UC Berkeley)
JevClaude CodeDevinClineCodex
Issue 010
Sep 21, 2026 · …
Jev goes fully open: Waitlist removed; same day open-sources Kev small model family and DocJev document classifier.Claude Code token-saving workshop: Official workshop goes viral in the Japanese community; Devin SWE-2 cloud free for a limited time.Harness Tax paper: UC Berkeley team finds similar success rates across harnesses but up to 5x cost difference — Pi is cheapest.
5× Cost difference across harnesses (UC Berkeley)
JevClaude CodeDevinClineCodex