AISpot

Codex 0.161.0 發布,GPT-6.1 Sol 成為預設模型

OpenAI Codex CLI releases10/7 17:01版本更新開發工具模型政策與安全

Codex 0.161.0 將 GPT-6.1 Sol 設為內建目錄與 Amazon Bedrock 目錄的預設模型。Amazon Bedrock 在相容模型上支援多智慧體 V2 與 Ultra 推理,Bedrock Mantle 也開始接受 AWS GovCloud 區域;終端會話可用 /mcp login <name> 登入 MCP 伺服器,語音對話可自選麥克風、揚聲器與輸入聲道並本地儲存偏好。

官方 changelog 確認 GPT-6.1 Sol 取代舊模型成為預設目錄模型,併為 Bedrock 新增多智慧體 V2 與 Ultra 推理支援,Daybreak 由預設開啟改為顯式 opt-in;對使用 Codex CLI 和在 Bedrock 上部署的開發者最直接。

原標題:0.161.0
閱讀原文

評分78 / 78(平均 78,門檻 60)
狀態精選

全文翻譯

新功能 - GPT-6.1 Sol 現在是捆綁目錄和 Amazon Bedrock 目錄中的預設模型。( #49318 , #49339 ) - Amazon Bedrock 在相容模型上支援多智慧體 V2 和 Ultra 推理;Bedrock Mantle 還接受 AWS GovCloud 區域。( #49345 , #49813 ) - 使用 /mcp login <name> 從活動終端會話登入 MCP 伺服器。( #49290 ) - 為語音對話選擇你的麥克風、揚聲器和麥克風輸入聲道,偏好設定會儲存在本地。( #49437 , #49836 ) - Daybreak 通過 --enable cli_daybreak 或 features.cli_daybreak=true 選擇啟用;僅 daybreak=true 不夠。預設情況下,控制元件和指示器隱藏,/daybreak 不可用,並且自動 Cyber 路由被省略——即使對於已儲存的 Daybreak 執行緒也是如此。已儲存的偏好設定保持不變。選擇啟用的路由需要符合條件的 ChatGPT 登入、OpenAI 提供商以及所公布的模型/計劃支援。( #49856 , #49858 , #49859 , #49861 , #51207 ) - 使用 codex exec --cyber-access-program 或 TypeScript SDK 的 cyberAccessProgram 選項,為每輪選擇 Cyber 訪問計劃。顯式 exec 覆蓋在 cli_daybreak 停用時仍然可用,並且不會更改已儲存的選擇。( #49939 , #51207 ) 錯誤修復 - 已批准的檔案系統提權現在可以授予更廣泛的寫入存取權限,同時保留被拒絕的讀取和網路限制。後臺任務保留其發起輪次的權限。( #49353 , #49880 ) - 顯式啟動權限在終端重新連線和新會話後仍然保留,而隱式客戶端設定不再覆蓋伺服器或已儲存執行緒的 Web 搜尋設定。( #49809 , #49799 ) - 提升權限的 Windows 終端會話可以使用嵌入式伺服器啟動,而沙盒化的 PowerShell 會保留受保護使用者配置檔案下的相對路徑。( #49855 , #49690 ) - 貼上檢測過期後,Enter 會正確提交緩衝輸入,包括在 Vim 插入模式下。( #49810 ) - 執行緒恢復包含最新已提交歷史。啟動時會更早檢測到可恢復的 SQLite 損壞,並將損壞的資料庫保留為備份。( #49599 , #49701 ) - Responses 重試和 WebSocket-to-HTTP 回退會遵循伺服器重試指引,減少過載期間的過早失敗。( #49441 ) 文件 - 身份驗證指南現在會考慮金鑰環儲存,而不是暗示憑據始終位於 auth.json 中。( #49361 ) 雜項 - 發布較舊的 alpha 或熱修復版本不再將 npm alpha 標籤向後移動。( #49704 ) 更新日誌 完整變更日誌:rust-v0.160.0...rust-v0.161.0 - #49246 在捆綁的 bwrap 測試中使用執行檔夾具複製 @felixxia-oai - #49257 允許 Guardian 在根上下文不完整時使用快取審批 @felixxia-oai - #49260 限制企業 MCP 認證,並在配置重新整理時故障關閉 @nicksteele-oai - #49261 保留 Windows sandbox runner 啟動錯誤 @zm-oai - #49262 跟蹤 turn 階段,並將已接受的輸入與 turn 關聯 @jif-oai - #49266 移除遠端 agent 訊息板客戶端 README @jif-oai - #49267 在多 agent 會話中支援遠端 agent 訊息板 @jif-oai - #49269 在配置過載期間保留 thread 覆蓋和雲策略有效性 @nicksteele-oai - #49275 將即時對話測試與 Responses 預熱連線隔離 @felixxia-oai - #49276 啟用企業 MCP 登入和帳戶範圍授權清理 @nicksteele-oai - #49277 避免 Guardian agent 訊息測試中的 turn 拆解競態 @felixxia-oai - #49280 將能力根限制為捕獲的 turn 環境 @miz-openai - #49286 將 exec-server 會話附加狀態建模為列舉 @jif-oai - #49290 向 TUI 新增 /mcp login <name> @nicksteele-oai - #49294 在審查和分類遙測中記錄 Guardian 上下文模式 @felixxia-oai - #49295 簡化配置指紋規範化 @jif-oai - #49297 為批次 thread 名稱查詢反向掃描會話索引 @jif-oai - #49300 每塊壓縮一次內聯隱藏標籤緩衝區 @jif-oai - #49305 解析 thread 名稱時批次讀取後設資料 @jif-oai - #49308 在沒有控制台的情況下執行管道式舊版 Windows 沙箱程序 @etraut-openai - #49312 當 Guardian 停止子 agent 時通知父 agent @jif-oai - #49316 在 CI 設定中將 taiki-e/install-action 升級到 v2.87.21 @jgershen-oai - #49318 將 GPT-6.1 Sol 新增為預設目錄模型 @andrewgu-oai - #49325 在錯誤 1056 時重試 Windows sandbox runner 登入一次 @zm-oai - #49330 在持續失敗期間保持遠端控制重連退避封頂 @yansenzhou-oai - #49332 立即清理已取消的 exec-server RPC 請求 @jif-oai - #49339 將 GPT-6.1 Sol 新增到 Bedrock 目錄並將其設為預設 @celia-oai - #49345 在 Amazon Bedrock 上啟用 multi-agent V2 和 Ultra reasoning @celia-oai - #49353 允許已批准的檔案系統提權,同時保留被拒絕的讀取 @viyatb-oai - #49357 貼上多行文字時繼續 Markdown 引用塊 @bc-openai - #49360 攜帶 shell 呼叫後設資料並報告執行器 PATH 目錄 @anp-oai - #49361 澄清跨認證 UI 和文件的憑據儲存措辭 @celia-oai - #49369 更新 Bedrock GPT-6 Sol 目錄測試以預期多代理 V2 @celia-oai - #49379 在發現期間編譯鉤子匹配器 @jif-oai - #49384 跟蹤憑據儲存結果並脫敏敏感錯誤 @celia-oai - #49388 修復帶斜槓字首的不透明 URI 的 Windows 路徑推斷 @dkovalenko-oai - #49389 序列化共享 Windows 沙箱帳戶的測試 @jgershen-oai - #49392 新增歸因的 MCP OAuth 憑據儲存遙測 @celia-oai - #49395 從 TUI 會話標題中移除隨機問候語 @etraut-openai - #49401 跨請求視窗保留即時工具呼叫後設資料 @ningyi-oai - #49403 為登入 shell 中的捆綁工具新增實驗性標誌 @anp-oai - #49406 支援使用 OpenAI API 金鑰的顯式網路訪問程式 @julee-oai - #49407 在環境資訊超時後恢復 exec-server 會話 @vivi - #49408 在記錄器重新整理測試中比較工具呼叫後設資料 @euroelessar - #49411 將 app-server 時間提供程式繫結到區域性變數 @euroelessar - #49414 從 SQLite 日誌中過濾優雅關閉防護和觸發跟蹤 @dkovalenko-oai - #49415 在協議除錯輸出中截斷輸入文字 @dkovalenko-oai - #49416 從多行 ANSI 警告中省略有效載荷 @dkovalenko-oai - #49424 使用正斜槓和混合斜槓推斷 Windows UNC 路徑 @dkovalenko-oai - #49425 按年齡和資料庫大小定期修剪診斷日誌 @dkovalenko-oai - #49426 預設對守護程序啟動的 app server 啟用分析 @bc-openai - #49432 跨認證更改保留引導發現 @cooper-oai - #49437 向 TUI 語音設定新增本地音訊裝置選擇 @bc-openai - #49441 在 Responses 重試和回退中遵循伺服器重試建議 @anp-oai - #49444 在反向 JSONL 掃描中使用 memrchr 查詢換行符 @btraut-openai - #49467 在登入 shell 啟動後恢復執行器工具路徑 @anp-oai - #49472 在 TUI 中使用伺服器權威權限 @etraut-openai - #49473 使用 rmcp SDK 進行企業管理令牌交換 @nicksteele-oai - #49475 在發出終端事件之前完成回合中止回撥 @euroelessar - #49478 在 ID-JAG 交換之前發現並驗證 MCP 授權伺服器 @nicksteele-oai - #49480 新增實驗性執行緒預測協議型別 @keyz - #49489 為分析批次之間的帳戶切換新增回歸覆蓋 @anp-oai - #49517 向 TUI 命令中心新增 fork 快捷方式 @etraut-openai - #49560 向多智慧體上下文新增一個可選加入的模型目錄 @jif-oai - #49564 在 TUI 中將選定的檔案路徑複製為純文字 @fcoury-oai - #49584 為 Guardian 審查跳過主機技能發現 @jif-oai - #49595 使嚴格的網路審批測試獨立於請求順序 @jif-oai - #49598 在模型歷史中持久化顯式的使用者目標編輯 @felixxia-oai - #49599 在恢復執行緒時返回權威的重放歷史 @felixxia-oai - #49600 在恢復執行緒時複用未更改的歷史快照 @felixxia-oai - #49624 對顯式遠端會話命令使用伺服器認證 @cooper-oai - #49642 允許受管要求停用 Windows MXC 沙箱 @iceweasel-oai - #49675 在大型輸入之前序列化 Responses 路由欄位 @jbeckwith-oai - #49678 在恢復問題答案時轉義命令草稿 @imac-oai - #49683 為應用內語音新增受管功能門控 @vishnu-oai - #49686 將遠端留言板通知投遞到活動輪次 @jif-oai - #49689 通過 OpenTelemetry 匯出技能呼叫事件 @jif-oai - #49690 在提升權限的 Windows 沙箱中保留 PowerShell 相對路徑 @johnl-oai - #49692 將壓縮的 rollout 片段搜尋保留在一個阻塞工作執行緒上 @npancha-openai - #49693 將執行緒歷史投影移入一個阻塞任務 @charliemarsh-oai - #49694 在可取消的阻塞工作執行緒上批次執行 rollout 列表掃描 @charliemarsh-oai - #49696 使 exec-server 檔案讀取在塊之間可取消 @charliemarsh-oai - #49701 在啟動期間檢測 SQLite 損壞並保留恢復備份 @dkovalenko-oai - #49702 重新命名 exec-server 檔案控制代碼管理識別符號 @anp-oai - #49704 防止 npm alpha dist-tags 向後移動 @imac-oai - #49706 升級 argument comment lint 工具鏈和 Dylint @tamird - #49708 將會話索引 I/O 移出非同步執行時執行緒 @charliemarsh-oai - #49710 使用型別化錯誤碼對 SQLite 損壞進行分類 @dkovalenko-oai - #49712 在令牌預算截斷中避免全字串掃描 @jif-oai - #49713 移除倉庫本地的 Codex 指南、技能和環境配置 @anp-oai - #49714 將 API 金鑰網路訪問程式與模型發現解耦 @julee-oai - #49715 向 TUI 新增帳戶安全設定提醒 @dennyku - #49778 為流式檔案寫入定義 exec-server 協議型別 @anp-oai - #49781 在 MCP 沙箱後設資料中包含環境的 MXC 後端 @zm-oai - #49782 清理失敗的 shell 快照捕獲的程序組 @jif-oai - #49783 在 TUI 中分叉時保留後臺執行緒請求 @etraut-openai - #49784 為瀏覽器註釋 API 新增 requirements 特性門控 @tepman-oai - #49785 在命名空的分頁執行緒時將其持久化 @rd-oai - #49786 澄清針對上下文目錄的 V2 spawn 模型覆蓋指導 @jif-oai - #49787 從 Bazel 核心測試資料中移除 AGENTS.md @andrewgu-oai - #49792 為 Guardian 非同步取樣新增保留對話支援 @felixxia-oai - #49793 為 Guardian v2 非同步分類新增對話模式 @felixxia-oai - #49795 避免 Guardian 分類器延續中出現重複的同步審查 @felixxia-oai - #49796 對 Guardian 保留上下文省略通知去重 @felixxia-oai - #49798 與 Arc 共享快取的 exec-server 環境資訊 @anp-oai - #49799 在 TUI 中保留伺服器網路搜尋設定 @etraut-openai - #49800 允許清理缺少執行緒的僅重放側對話 @bc-openai - #49801 更新用於引數註釋 lint 檢查的 Rust 工具鏈 action @andrewgu-oai - #49804 在 TUI 快捷鍵提示中使用平台特定的修飾鍵標簽 @bc-openai - #49805 為 exec-server 客戶端新增能力門控的可寫檔案流 @anp-oai - #49806 在 app-server 協議中接受未知的 Codex 錯誤變體 @anp-oai - #49807 預設啟用 API 金鑰模型發現 @andrewgu-oai - #49809 在 TUI 會話和重連之間保留本地啟動權限 @etraut-openai - #49810 在處理 Enter 之前重新整理過期的貼上突發 @bc-openai - #49811 在 exec-server 中處理不支援的 fs/writeBlock 請求 @anp-oai - #49812 將影子技能排名移出回合準備路徑 @vkg-oai - #49813 支援 Amazon Bedrock Mantle 的 AWS GovCloud 區域 @alexsong-oai - #49814 為本地代理樹新增協調關閉 @owenlin0 - #49816 從 TUI 中移除瀏覽器開啟成功訊息 @bc-openai - #49817 新增建議性的 Bedrock GovCloud requirements 檢查 @alexsong-oai - #49818 為沙盒檔案開啟使用專用引數 @anp-oai - #49819 在 cwd 被刪除後恢復守護程序啟動和更新器重新執行 @etraut-openai - #49822 在 agents 概覽權限測試中將 resume future 裝箱 @sayan-oai - #49835 澄清 TUI 中服務層級預設儲存錯誤 @etraut-openai - #49836 允許為語音對話選擇麥克風通道 @bc-openai - #49843 保留守護程序診斷資訊,並在報告中包含更新器日誌 @etraut-openai - #49846 為每個回合捕獲主機提供的擴充套件資料 @sayan-oai - #49847 在渲染上下文的同時持久化世界狀態快照 @pakrym-oai - #49850 在專用工作目錄中啟動 Windows 守護程序子程序 @etraut-openai - #49852 改進報告附件失敗的診斷資訊 @dkovalenko-oai - #49855 對提權 Windows TUI 會話使用嵌入式模式 @etraut-openai - #49856 在 codex exec 中支援 Daybreak 選擇 @etraut-openai - #49857 使用模型目錄選擇 TUI 網路拒絕指導 @etraut-openai - #49858 向 TUI 新增持久化的 /daybreak 開關 @etraut-openai - #49859 在 TUI 續接和後臺任務中遵循 Daybreak 設定 @etraut-openai - #49861 將 Daybreak 狀態新增到狀態行和終端標題 @etraut-openai - #49867 更新提權啟動警告快照以使用 ⌃o 進行復制 @etraut-openai - #49874 將用量和額度連結指向 ChatGPT 設定 @etraut-openai - #49875 將 TUI 啟動展示與執行配置解耦 @etraut-openai - #49876 從 TUI 中移除個性化管道 @etraut-openai - #49880 將權限授予繫結到發起的輪次 @anp-oai - #49894 一起返回世界狀態快照和上下文更新 @pakrym-oai - #49898 將擴充套件檔案系統訪問範圍限定為回撥權限 @anp-oai - #49910 保留無效 TUI 鍵位繫結的驗證錯誤 @etraut-openai - #49912 在臨時結構化執行緒中遵循審批策略 @etraut-openai - #49939 向 exec 和 SDK 新增按輪次的 Cyber 訪問程式選擇 @mldangelo-oai - #49946 防止過期的檔案搜尋結果被標記為新查詢 @jif-oai - #49951 在 Guardian 傳送者審查中包含前置助手上下文 @jif-oai - #49956 為 MCP 鉤子引數展開快取佔位符正規表示式 @jif-oai - #49959 測試會話索引執行緒名稱的追加和移除 @jif-oai - #49972 在 exec-server 輸出塊之間共享位元組緩衝區 @jif-oai - #49987 新增可續期的 EMA HTTP 身份驗證和憑證版本管理 @nicksteele-oai - #49993 在保留上下文變化時保留非同步 Guardian 歷史字首 @felixxia-oai

由 AI 翻譯,以原文為準。

原文
New Features GPT-6.1 Sol is now the default model in the bundled and Amazon Bedrock catalogs. ( #49318 , #49339 ) Amazon Bedrock supports multi-agent V2 and Ultra reasoning on compatible models; Bedrock Mantle also accepts AWS GovCloud regions. ( #49345 , #49813 ) Sign in to MCP servers from an active terminal session with /mcp login <name> . ( #49290 ) Choose your microphone, speaker, and microphone input channels for voice conversations, with preferences saved locally. ( #49437 , #49836 ) Daybreak is opt-in through --enable cli_daybreak or features.cli_daybreak=true ; daybreak=true alone is insufficient. By default, controls and indicators are hidden, /daybreak is unavailable, and automatic Cyber routing is omitted—even for saved Daybreak threads. Saved preferences remain intact. Opt-in routing requires eligible ChatGPT sign-in, the OpenAI provider, and advertised model/program support. ( #49856 , #49858 , #49859 , #49861 , #51207 ) Select a Cyber access program per turn with codex exec --cyber-access-program or the TypeScript SDK’s cyberAccessProgram option. The explicit exec override remains available with cli_daybreak disabled and leaves the saved choice unchanged. ( #49939 , #51207 ) Bug Fixes Approved filesystem escalation can now grant broader write access while preserving denied reads and network restrictions. Background tasks retain their originating turn’s permissions. ( #49353 , #49880 ) Explicit launch permissions survive terminal reconnects and new sessions, while implicit client settings no longer overwrite server or saved-thread web-search settings. ( #49809 , #49799 ) Elevated Windows terminal sessions can start using an embedded server, and sandboxed PowerShell preserves relative paths beneath protected user profiles. ( #49855 , #49690 ) Enter correctly submits buffered input after paste detection expires, including in Vim insert mode. ( #49810 ) Thread resume includes the latest committed history. Startup detects recoverable SQLite corruption earlier and preserves damaged databases as backups. ( #49599 , #49701 ) Responses retries and WebSocket-to-HTTP fallback honor server retry guidance, reducing premature failures during overload. ( #49441 ) Documentation Authentication guidance now accounts for keyring storage instead of implying credentials always reside in auth.json . ( #49361 ) Chores Publishing an older alpha or hotfix no longer moves npm alpha tags backward. ( #49704 ) Changelog Full Changelog: rust-v0.160.0...rust-v0.161.0 #49246 Use executable fixture copying in the bundled bwrap test @felixxia-oai #49257 Allow Guardian cached approvals with incomplete root context @felixxia-oai #49260 Restrict enterprise MCP auth and fail closed on config refresh @nicksteele-oai #49261 Preserve Windows sandbox runner launch errors @zm-oai #49262 Trace turn phases and correlate accepted input with turns @jif-oai #49266 Remove the remote agent message board client README @jif-oai #49267 Support remote agent message boards in multi-agent sessions @jif-oai #49269 Preserve thread overrides and cloud policy validity during config reloads @nicksteele-oai #49275 Isolate realtime conversation tests from Responses prewarm connections @felixxia-oai #49276 Enable enterprise MCP sign-in and account-scoped grant cleanup @nicksteele-oai #49277 Avoid a turn teardown race in the Guardian agent message test @felixxia-oai #49280 Restrict capability roots to captured turn environments @miz-openai #49286 Model exec-server session attachment state as an enum @jif-oai #49290 Add /mcp login <name> to the TUI @nicksteele-oai #49294 Record Guardian context mode in review and classification telemetry @felixxia-oai #49295 Simplify configuration fingerprint canonicalization @jif-oai #49297 Scan the session index backwards for batch thread name lookups @jif-oai #49300 Compact the inline hidden tag buffer once per chunk @jif-oai #49305 Batch metadata reads when resolving thread names @jif-oai #49308 Run piped legacy Windows sandbox processes without a console @etraut-openai #49312 Notify parent agents when Guardian stops a subagent @jif-oai #49316 Bump taiki-e/install-action to v2.87.21 in CI setup @jgershen-oai #49318 Add GPT-6.1 Sol as the default catalog model @andrewgu-oai #49325 Retry Windows sandbox runner logon once on error 1056 @zm-oai #49330 Keep remote control reconnect backoff capped during sustained failures @yansenzhou-oai #49332 Clean up canceled exec-server RPC requests immediately @jif-oai #49339 Add GPT-6.1 Sol to Bedrock catalogs and make it the default @celia-oai #49345 Enable multi-agent V2 and Ultra reasoning on Amazon Bedrock @celia-oai #49353 Allow approved filesystem escalation while preserving denied reads @viyatb-oai #49357 Continue Markdown blockquotes when pasting multiline text @bc-openai #49360 Carry shell invocation metadata and report executor PATH directories @anp-oai #49361 Clarify credential storage wording across authentication UI and docs @celia-oai #49369 Update Bedrock GPT-6 Sol catalog tests to expect multi-agent V2 @celia-oai #49379 Compile hook matchers during discovery @jif-oai #49384 Track credential storage outcomes and redact sensitive errors @celia-oai #49388 Fix Windows path inference for opaque URIs with slash prefixes @dkovalenko-oai #49389 Serialize tests that share Windows sandbox accounts @jgershen-oai #49392 Add attributed MCP OAuth credential storage telemetry @celia-oai #49395 Remove randomized greetings from TUI session headers @etraut-openai #49401 Preserve live tool-call metadata across request windows @ningyi-oai #49403 Add an experimental flag for bundled tools in login shells @anp-oai #49406 Support explicit cyber access programs with OpenAI API keys @julee-oai #49407 Recover exec-server sessions after environment info timeouts @vivi #49408 Compare tool call metadata in the recorder refresh test @euroelessar #49411 Bind the app-server time provider to a local variable @euroelessar #49414 Filter graceful shutdown guard and trigger traces from SQLite logs @dkovalenko-oai #49415 Truncate input text in protocol debug output @dkovalenko-oai #49416 Omit payloads from multiline ANSI warnings @dkovalenko-oai #49424 Infer Windows UNC paths with forward and mixed slashes @dkovalenko-oai #49425 Prune diagnostic logs periodically by age and database size @dkovalenko-oai #49426 Enable analytics by default for daemon-launched app servers @bc-openai #49432 Preserve bootstrap discovery across authentication changes @cooper-oai #49437 Add local audio device selection to TUI voice settings @bc-openai #49441 Honor server retry advice across Responses retries and fallback @anp-oai #49444 Use memrchr to find newlines in reverse JSONL scans @btraut-openai #49467 Restore executor tool paths after login shell startup @anp-oai #49472 Use server-authoritative permissions in the TUI @etraut-openai #49473 Use the rmcp SDK for enterprise-managed token exchanges @nicksteele-oai #49475 Complete turn abort callbacks before emitting terminal events @euroelessar #49478 Discover and validate MCP authorization servers before ID-JAG exchange @nicksteele-oai #49480 Add experimental thread prediction protocol types @keyz #49489 Add regression coverage for account switches between analytics batches @anp-oai #49517 Add a fork shortcut to the TUI command center @etraut-openai #49560 Add an opt-in model catalog to multi-agent context @jif-oai #49564 Copy selected file paths as plain text in the TUI @fcoury-oai #49584 Skip host skill discovery for Guardian reviews @jif-oai #49595 Make the strict network approval test independent of request order @jif-oai #49598 Persist explicit user goal edits in model history @felixxia-oai #49599 Return authoritative replay history when resuming a thread @felixxia-oai #49600 Reuse unchanged history snapshots when resuming threads @felixxia-oai #49624 Use server authentication for explicit remote session commands @cooper-oai #49642 Allow managed requirements to disable the Windows MXC sandbox @iceweasel-oai #49675 Serialize Responses routing fields before large inputs @jbeckwith-oai #49678 Escape command drafts when recovering question answers @imac-oai #49683 Add a managed feature gate for in-app voice @vishnu-oai #49686 Deliver remote message board notifications to active turns @jif-oai #49689 Export skill invocation events through OpenTelemetry @jif-oai #49690 Preserve PowerShell relative paths in the elevated Windows sandbox @johnl-oai #49692 Keep compressed rollout snippet searches on one blocking worker @npancha-openai #49693 Move thread history projection into one blocking task @charliemarsh-oai #49694 Batch rollout listing scans on cancellable blocking workers @charliemarsh-oai #49696 Make exec-server file reads cancellable between chunks @charliemarsh-oai #49701 Detect SQLite corruption during startup and preserve recovery backups @dkovalenko-oai #49702 Rename exec-server file handle management identifiers @anp-oai #49704 Prevent npm alpha dist-tags from moving backward @imac-oai #49706 Upgrade the argument comment lint toolchain and Dylint @tamird #49708 Move session index I/O off async runtime threads @charliemarsh-oai #49710 Classify SQLite corruption using typed error codes @dkovalenko-oai #49712 Avoid full-string scans in token-budget truncation @jif-oai #49713 Remove repository-local Codex guidance, skills, and environment config @anp-oai #49714 Decouple API-key cyber access programs from model discovery @julee-oai #49715 Add account security setup reminders to the TUI @dennyku #49778 Define exec-server protocol types for streamed file writes @anp-oai #49781 Include the environment's MXC backend in MCP sandbox metadata @zm-oai #49782 Clean up process groups for failed shell snapshot captures @jif-oai #49783 Preserve background thread requests when forking in the TUI @etraut-openai #49784 Add a requirements feature gate for the browser annotation API @tepman-oai #49785 Persist empty paginated threads when naming them @rd-oai #49786 Clarify V2 spawn model override guidance for context catalogs @jif-oai #49787 Remove AGENTS.md from Bazel core test data @andrewgu-oai #49792 Add retained conversation support to Guardian async sampling @felixxia-oai #49793 Add conversation mode to Guardian v2 async classification @felixxia-oai #49795 Avoid duplicate sync reviews in Guardian classifier continuations @felixxia-oai #49796 Deduplicate Guardian retained-context omission notices @felixxia-oai #49798 Share cached exec-server environment info with Arc @anp-oai #49799 Preserve server web-search settings in the TUI @etraut-openai #49800 Allow cleanup of replay-only side conversations with missing threads @bc-openai #49801 Update the Rust toolchain action for argument-comment linting @andrewgu-oai #49804 Use platform-specific modifier labels in TUI shortcut hints @bc-openai #49805 Add capability-gated writable file streams to the exec-server client @anp-oai #49806 Accept unknown Codex error variants in the app-server protocol @anp-oai #49807 Enable API-key model discovery by default @andrewgu-oai #49809 Preserve local launch permissions across TUI sessions and reconnects @etraut-openai #49810 Flush expired paste bursts before handling Enter @bc-openai #49811 Handle unsupported fs/writeBlock requests in exec-server @anp-oai #49812 Move shadow skill ranking off the turn preparation path @vkg-oai #49813 Support AWS GovCloud regions for Amazon Bedrock Mantle @alexsong-oai #49814 Add coordinated shutdown for local agent trees @owenlin0 #49816 Remove browser-open success messages from the TUI @bc-openai #49817 Add an advisory Bedrock GovCloud requirements check @alexsong-oai #49818 Use dedicated parameters for sandboxed file opens @anp-oai #49819 Recover daemon startup and updater re-exec after cwd deletion @etraut-openai #49822 Box the resume future in the agents overview permissions test @sayan-oai #49835 Clarify service tier default save errors in the TUI @etraut-openai #49836 Allow microphone channel selection for voice conversations @bc-openai #49843 Preserve daemon diagnostics and include updater logs in reports @etraut-openai #49846 Capture host-supplied extension data for each turn @sayan-oai #49847 Persist world-state snapshots alongside rendered context @pakrym-oai #49850 Launch Windows daemon children in a dedicated working directory @etraut-openai #49852 Improve diagnostics for report attachment failures @dkovalenko-oai #49855 Use embedded mode for elevated Windows TUI sessions @etraut-openai #49856 Support Daybreak selection in codex exec @etraut-openai #49857 Use the model catalog to select TUI cyber refusal guidance @etraut-openai #49858 Add a persistent /daybreak toggle to the TUI @etraut-openai #49859 Honor Daybreak settings in TUI continuations and background tasks @etraut-openai #49861 Add Daybreak state to the status line and terminal title @etraut-openai #49867 Update elevated-launch warning snapshot to use ⌃o for copy @etraut-openai #49874 Point usage and credit links to ChatGPT settings @etraut-openai #49875 Decouple TUI startup presentation from execution configuration @etraut-openai #49876 Remove personality plumbing from the TUI @etraut-openai #49880 Bind permission grants to the originating turn @anp-oai #49894 Return world-state snapshots and context updates together @pakrym-oai #49898 Scope extension filesystem access to callback permissions @anp-oai #49910 Preserve validation errors for invalid TUI keybindings @etraut-openai #49912 Respect approval policies in temporary structured threads @etraut-openai #49939 Add per-turn Cyber access program selection to exec and the SDK @mldangelo-oai #49946 Prevent stale file search results from being labeled with a new query @jif-oai #49951 Include preceding assistant context in Guardian sender reviews @jif-oai #49956 Cache the placeholder regex for MCP hook argument expansion @jif-oai #49959 Test session index thread-name append and removal @jif-oai #49972 Share byte buffers across exec-server output chunks @jif-oai #49987 Add renewable EMA HTTP authentication and credential versioning @nicksteele-oai #49993 Preserve the async Guardian history prefix as retained context changes @felixxia-oai

相關報導

OpenAI Codex CLI releases10/5 19:31AI 評分41

OpenAI Codex 發布 0.160.1 修復遠端 MCP 環境變數

OpenAI 的 Codex 倉庫發布 0.160.1,這是面向 0.160 分支的補丁版本。修復內容為:在用顯式配置的遠端環境變數啟動遠端 stdio MCP 伺服器時,保留 SYSTEMROOT、TEMP 和 TMP,使 Unix 主機能夠沿用 Windows 執行器的啟動環境。該改動以 #51121 回移植到 0.160 分支,發布時 main 分支已比此版本多出 287 次提交。

量子位● 精選10/6 07:45AI 評分89

OpenAI 瘋狂 28 天首日:GPT-6 提速 50%,網友實測未達標

OpenAI Codex 負責人 Tibo 的 28 天計劃首日,官方宣布 GPT-6 Astra 與 GPT-6.1 Sol 預設推理速度提升約 50%、達 50TPS,覆蓋官方訂閱及 OpenCode、Amp 等第三方接入,但網友實測僅約 35TPS。同期 OpenAI 在 ChatGPT 影像生成中測試視覺廣告,並將在幾週內為歐盟地區 ChatGPT 和 Codex 生成的文字加隱形水印。

Latent Space● 精選9/30 23:23AI 評分82

OpenAI 發布 Dots、GPT-6.1 Sol 與 Agents API

OpenAI 在 DevDay 上發布個人助理產品 Dots、新模型 GPT-6.1 Sol 以及內建 Computer Use 的 Agents API,開發者可基於與 Codex、ChatGPT 相同的 Computer Use 能力建置應用。GPT-6.1 Sol 的 Computer Use 成本約為 Astra 的七分之一。播客還討論了非同步工具呼叫、UltraFast 推理、Decisions API、提示快取與上下文壓縮等新開發者棧。

OpenAI Codex changelog● 精選10/7 01:00AI 評分61

ChatGPT iOS 版更新:支援直接開啟 Codex 任務連結

ChatGPT for iOS 發布 1.2026.272 版本,新增回覆內聯頁面預覽,並支援在 iOS 上直接開啟 Codex 任務連結。審批選擇器重新設計,權限選項更清晰;外掛入口統一為輸入框加號選單裡的 Plugins,長會話流式輸出更快、卡頓更少。新任務在電腦重連或離線時仍保留所選裝置,worktree 支援改進以匹配桌面端檢出,同時修復了連線後任務列表為空、狀態更新顯示為待回答問題等問題。