@bitkyc08/opencodex 2.6.31-preview.20260707 → 2.6.32

package/README.ko.md

@@ -139,6 +139,15 @@ codex -m "ollama/llama3" "이 함수를 리팩터링해 줘"
 
 라우팅된 모델은 **Codex App** 모델 선택기에도 모델별 reasoning effort 컨트롤과 함께 나타납니다:
 
+현재 Codex 빌드는 모델이 광고하는 경우 `low`, `medium`, `high`, `xhigh`, `max` reasoning 컨트롤을
+노출할 수 있습니다. opencodex는 프로바이더 config가 명시적으로 alias를 지정하지 않는 한 `xhigh`와
+`max`를 서로 다른 단계로 유지합니다.
+
+GPT-5.6 Sol/Terra/Luna는 OpenAI API key 및 OpenRouter preset에서 rollout-ready catalog 항목으로
+seed됩니다(`gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`; OpenRouter는 `openai/...` 사용). 실제
+사용 가능 여부는 upstream preview gate를 따르며, opencodex는 계정/프로바이더가 제공할 때 쓸
+routing/catalog metadata를 준비해 둡니다.
+
 <p align="center">
   <img src="assets/codex-app-picker.png" alt="opencodex 라우팅 모델을 reasoning effort 선택기와 함께 보여주는 Codex App" width="480">
 </p>
@@ -164,8 +173,9 @@ opencodex는 두 가지 동작을 분리해서 유지합니다:
 - **한 번 로그인하면 API 키는 생략.** xAI, Anthropic, Kimi는 OAuth를 지원하므로 기존 계정으로 인증할 수 있고 토큰은 자동 갱신됩니다. 또는 `codex login`을 forward 하거나, API 키를 붙여넣거나, `${ENV_VAR}` 참조를 쓸 수 있습니다 — 선택은 자유입니다.
 - **Codex가 동작하는 모든 곳에서.** Codex CLI, TUI, App, SDK에 자동으로 주입됩니다. 라우팅된 모델이 네이티브 모델처럼 Codex 모델 선택기에 나타납니다.
 - **알맞은 모델에 위임.** 대시보드나 config에서 최대 5개의 라우팅/네이티브 모델을 Codex 서브에이전트 선택기에 노출해, 복잡한 작업은 reasoning 모델로, 빠른 작업은 저렴한 모델로 보낼 수 있습니다.
+- **프리뷰 게이트된 OpenAI rollout에 대비.** GPT-5.6 Sol/Terra/Luna fallback 항목은 OpenAI API key와 OpenRouter route용으로 준비되어 있으며, upstream access가 있을 때 `max` reasoning과 372k usable-context metadata를 사용합니다.
 - **어떤 모델에도 초능력을.** OpenAI가 아닌 모델도 ChatGPT 로그인 위에서 도는 `gpt-5.4-mini` sidecar로 실제 웹 검색과 이미지 이해를 사용합니다.
-- **무슨 일이 일어나는지 보이게.** 웹 대시보드가 프로바이더, OAuth 상태, 모델 선택, 실시간 요청 로그를 보여줍니다 — 왜 요청이 실패했는지 더는 추측하지 않아도 됩니다.
+- **무슨 일이 일어나는지 보이게.** 웹 대시보드가 프로바이더, OAuth 상태, 모델 선택, upstream이 보고한 cached/cache-write 토큰 수를 포함한 실시간 요청 로그를 보여줍니다 — 왜 요청이 실패했는지 더는 추측하지 않아도 됩니다.
 - **백그라운드 실행.** 시스템 서비스(launchd / systemd / Task Scheduler)로 설치하면 부팅 시 자동 시작되어 신경 쓸 필요가 없습니다.
 - **깔끔한 종료, 잔여물 제로.** `ocx stop`(또는 대시보드의 Stop 버튼)은 프록시를 종료하고, 설치된 백그라운드 서비스를 멈추며, Codex를 원래 설정으로 복원합니다. 이후 `codex`는 잔여 설정이나 좀비 프로세스 없이 이전과 똑같이 동작합니다.
 
@@ -266,8 +276,10 @@ opencodex는 `config.json.invalid-<timestamp>`로 백업하고 경고를 출력
 프로바이더 항목은 라우팅 카탈로그 메타데이터도 함께 지정할 수 있습니다. `contextWindow`는 프로바이더
 전체에 적용되는 Codex 노출용 컨텍스트 상한, `modelContextWindows`는 모델별 상한,
 `modelInputModalities`는 `["text"]`나 `["text", "image"]` 같은 모델별 입력 힌트입니다. 이 값들은 라이브
-`/models` 메타데이터를 상한으로 제한할 뿐, 더 작은 라이브 컨텍스트를 늘리지는 않습니다. 전체 필드는
-설정 레퍼런스를 참고하세요.
+`/models` 메타데이터를 상한으로 제한할 뿐, 더 작은 라이브 컨텍스트를 늘리지는 않습니다. 번들된 GPT-5.6
+Sol/Terra/Luna fallback metadata는 OpenAI API key와 OpenRouter catalog 항목에 372,000 토큰 usable
+context window를 사용하며, upstream preview access를 우회하지 않습니다. 전체 필드는 설정 레퍼런스를
+참고하세요.
 
 > **Z.AI 경유 GLM-5.2 1M 컨텍스트:** `openai-chat` adapter에서는 `glm-5.2`와 `glm-5.2[1m]`이 모두
 > 동작합니다 — opencodex가 요청 전에 끝의 `[1m]` 접미사를 제거하기 때문입니다(OpenAI 호환 엔드포인트는

package/README.md

@@ -141,6 +141,15 @@ When you omit the `provider/` prefix, opencodex routes to the default provider
 
 Routed models also appear in the **Codex App** model picker with per-model reasoning effort controls:
 
+Current Codex builds can expose `low`, `medium`, `high`, `xhigh`, and `max` reasoning controls when a
+model advertises them. opencodex keeps `xhigh` and `max` distinct unless a provider config explicitly
+maps one to the other.
+
+GPT-5.6 Sol/Terra/Luna are seeded as rollout-ready catalog entries for the OpenAI API key and
+OpenRouter presets (`gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`; OpenRouter uses
+`openai/...`). They remain preview-gated by upstream availability; opencodex only prepares the
+routing and catalog metadata for accounts and providers that can serve them.
+
 <p align="center">
   <img src="assets/codex-app-picker.png" alt="Codex App showing opencodex routed models with reasoning effort picker" width="480">
 </p>
@@ -170,8 +179,9 @@ next Codex session. opencodex keeps two separate behaviors:
 - **Works everywhere Codex does.** Injects into Codex CLI, TUI, App, and SDK automatically. Routed models show up in Codex's model picker just like native ones.
 - **History-safe injection.** On local installs the proxy points Codex's own built-in `openai` provider at itself via a single `openai_base_url` line — new threads keep their native provider tag, so ongoing chat history is never remapped and an unclean shutdown can't hide it. (Threads re-tagged by older versions are migrated back once on the first start; remote/LAN binds use a dedicated provider entry instead, since they need an API-key header.)
 - **Delegate to the right model.** Feature up to five routed or native models in Codex's subagent picker from the dashboard or config — route complex tasks to a reasoning model, fast tasks to a cheap one.
+- **Prepare for preview-gated OpenAI rollouts.** GPT-5.6 Sol/Terra/Luna fallback entries are ready for OpenAI API key and OpenRouter routes, with `max` reasoning and 372k usable-context metadata when upstream access is available.
 - **Give any model superpowers.** Non-OpenAI models get real web search and image understanding via a `gpt-5.4-mini` sidecar over your ChatGPT login.
-- **See what's happening.** The web dashboard shows providers, OAuth status, model selection, and a live request log — no more guessing why a request failed.
+- **See what's happening.** The web dashboard shows providers, OAuth status, model selection, and a live request log, including cached/cache-write token counts when upstream reports them — no more guessing why a request failed.
 - **Runs in the background.** Install as a system service (launchd / systemd / Task Scheduler) and forget about it. The proxy starts on boot and stays out of your way.
 - **Clean exit, zero residue.** `ocx stop` (or the dashboard's Stop button) shuts down the proxy, stops the background service if one is installed, and restores Codex to its original configuration. Plain `codex` works exactly as it did before — no leftover config, no orphaned processes.
 
@@ -286,7 +296,9 @@ Provider entries can also annotate routed catalog metadata. Use `contextWindow`
 Codex-visible context cap, `modelContextWindows` for model-specific caps, and
 `modelInputModalities` for model-specific catalog input hints such as `["text"]` or
 `["text", "image"]`. Context values cap live `/models` metadata; they never raise a smaller live
-context window. See the configuration reference for the full field list.
+context window. The bundled GPT-5.6 Sol/Terra/Luna fallback metadata uses a 372,000-token usable
+context window for OpenAI API key and OpenRouter catalog entries; it does not bypass upstream preview
+access. See the configuration reference for the full field list.
 
 > **GLM-5.2 1M context via Z.AI:** through the `openai-chat` adapter, both `glm-5.2`
 > and `glm-5.2[1m]` work — opencodex strips the trailing `[1m]` suffix before

package/README.zh-CN.md

@@ -100,8 +100,9 @@ npm install -g @bitkyc08/opencodex   # 不要加 --ignore-scripts、--omit=optio
 - **登录一次，免填 API key。** xAI、Anthropic、Kimi 支持 OAuth，可用现有账户认证，token 自动刷新。也可以转发 `codex login`、粘贴 API key，或使用 `${ENV_VAR}` 引用 —— 随你选择。
 - **Codex 在哪里能用，它就在哪里能用。** 自动注入 Codex CLI、TUI、App 和 SDK。路由模型像原生模型一样出现在 Codex 的模型选择器里。
 - **委派给合适的模型。** 在仪表盘或 config 中把最多 5 个路由/原生模型放进 Codex 的 subagent 选择器 —— 复杂任务交给 reasoning 模型，快速任务交给便宜模型。
+- **为 preview-gated OpenAI rollout 做好准备。** GPT-5.6 Sol/Terra/Luna fallback 条目已可用于 OpenAI API key 和 OpenRouter 路由；当上游访问可用时，会带有 `max` reasoning 和 372k usable-context 元数据。
 - **给任意模型超能力。** 非 OpenAI 模型也能通过你的 ChatGPT 登录上运行的 `gpt-5.4-mini` sidecar 获得真正的网页搜索和图片理解。
-- **看清正在发生什么。** Web 仪表盘展示 provider、OAuth 状态、模型选择和实时请求日志 —— 不必再猜测请求为何失败。
+- **看清正在发生什么。** Web 仪表盘展示 provider、OAuth 状态、模型选择和实时请求日志；当上游返回时，也会包含 cached/cache-write token 计数 —— 不必再猜测请求为何失败。
 - **后台运行。** 安装为系统服务（launchd / systemd / Task Scheduler）后开机自启，无需操心。
 - **干净退出，零残留。** `ocx stop`（或仪表盘的 Stop 按钮）会关闭代理、停止已安装的后台服务，并将 Codex 恢复为原始配置。之后 `codex` 就像从未安装过 opencodex 一样工作 —— 无残留配置，无僵尸进程。
 
@@ -147,6 +148,13 @@ codex -m "ollama/llama3" "重构这个函数"
 
 路由模型也会出现在 **Codex App** 模型选择器中，并带有按模型的 reasoning effort 控制：
 
+当前 Codex 构建在模型声明支持时可显示 `low`、`medium`、`high`、`xhigh` 和 `max` reasoning 控制。
+除非 provider config 明确设置 alias，opencodex 会把 `xhigh` 与 `max` 保持为不同档位。
+
+GPT-5.6 Sol/Terra/Luna 已在 OpenAI API key 和 OpenRouter 预设中作为 rollout-ready 目录条目预置
+（`gpt-5.6-sol`、`gpt-5.6-terra`、`gpt-5.6-luna`；OpenRouter 使用 `openai/...`）。可用性仍受上游
+preview gate 限制；opencodex 只是准备好你的账户/provider 可访问时所需的路由和目录元数据。
+
 <p align="center">
   <img src="assets/codex-app-picker.png" alt="Codex App 展示 opencodex 路由模型及 reasoning effort 选择器" width="480">
 </p>
@@ -256,8 +264,9 @@ npm uninstall -g @bitkyc08/opencodex
 
 provider 条目还可以标注路由目录元数据。`contextWindow` 设置 provider 级别、对 Codex 可见的上下文上限，
 `modelContextWindows` 设置按模型的上限，`modelInputModalities` 设置按模型的目录输入提示，例如 `["text"]`
-或 `["text", "image"]`。这些值只会对实时 `/models` 元数据设上限，绝不会抬高更小的实时上下文窗口。完整字段
-参阅配置参考。
+或 `["text", "image"]`。这些值只会对实时 `/models` 元数据设上限，绝不会抬高更小的实时上下文窗口。内置
+GPT-5.6 Sol/Terra/Luna fallback 元数据会为 OpenAI API key 和 OpenRouter 目录条目使用 372,000 token 的
+usable context window；它不会绕过上游 preview access。完整字段参阅配置参考。
 
 > **通过 Z.AI 使用 GLM-5.2 1M 上下文：** 在 `openai-chat` adapter 下，`glm-5.2` 和 `glm-5.2[1m]` 都可用 ——
 > opencodex 会在发送请求前剥离末尾的 `[1m]` 后缀，因为 OpenAI 兼容端点会拒绝带方括号的 id（Z.AI 400 code

package/gui/dist/assets/{index-CWujz83O.js → index-D_JZzI0r.js}

@@ -12,4 +12,4 @@ Error generating stack: `+e.message+`
   -d '{
     "model": "gpt-5.4",
     "input": "Hello, world!"
-  }'`})]})]})}var xt=`opencodex-api-token`,St=!1,Ct=null;function wt(e){try{let t=e instanceof Request?e.url:String(e);return new URL(t,window.location.href).pathname}catch{return null}}function Tt(e){let t=wt(e);return!!t&&(t.startsWith(`/api/`)||t.startsWith(`/v1/`))}function Et(){try{return sessionStorage.getItem(xt)?.trim()||null}catch{return null}}function Dt(e){try{sessionStorage.setItem(xt,e)}catch{}}function Ot(){try{sessionStorage.removeItem(xt)}catch{}}function kt(e,t,n){let r=new Headers(t?.headers??(e instanceof Request?e.headers:void 0));return r.set(`X-OpenCodex-API-Key`,n),e instanceof Request?[new Request(e,{headers:r}),t?{...t,headers:r}:void 0]:[e,{...t,headers:r}]}async function At(){return Ct||(Ct=Promise.resolve().then(()=>window.prompt(`OpenCodex API token`)?.trim()||null).finally(()=>{Ct=null}),Ct)}function jt(){if(St)return;St=!0;let e=window.fetch.bind(window);window.fetch=async(t,n)=>{if(!Tt(t))return e(t,n);let r=Et(),[i,a]=r?kt(t,n,r):[t,n],o=await e(i,a);if(o.status!==401)return o;r&&Ot();let s=await At();if(!s)return o;Dt(s);let[c,l]=kt(t,n,s),u=await e(c,l);return u.status===401&&Ot(),u}}jt();var Mt=new Set([`dashboard`,`providers`,`models`,`subagents`,`logs`,`usage`,`codex-auth`,`api`]);function Nt(){let e=location.hash.replace(/^#\/?/,``);return Mt.has(e)?e:`dashboard`}var Pt=``,Ft=`ocx-theme`,It=[{id:`dashboard`,tkey:`nav.dashboard`,Icon:w},{id:`providers`,tkey:`nav.providers`,Icon:T},{id:`models`,tkey:`nav.models`,Icon:E},{id:`subagents`,tkey:`nav.subagents`,Icon:ee},{id:`logs`,tkey:`nav.logs`,Icon:D},{id:`usage`,tkey:`nav.usage`,Icon:te},{id:`codex-auth`,tkey:`nav.codexAuth`,Icon:L},{id:`api`,tkey:`nav.api`,Icon:de}],Lt={light:ce,dark:le,system:ue},Rt={light:`theme.light`,dark:`theme.dark`,system:`theme.system`};function zt(e){if(!e||typeof e!=`object`||!(`version`in e))return null;let t=e.version;return typeof t==`string`&&t.length>0?t:null}function Bt(){let e=localStorage.getItem(Ft);return e===`light`||e===`dark`?e:`system`}function Vt(){let[e,t]=(0,_.useState)(Nt),n=e=>{location.hash=e,t(e)},[r,i]=(0,_.useState)(Bt),[a,o]=(0,_.useState)(null),{locale:s,setLocale:c}=be(),l=xe();(0,_.useEffect)(()=>{let e=()=>t(Nt());return window.addEventListener(`hashchange`,e),()=>window.removeEventListener(`hashchange`,e)},[]),(0,_.useEffect)(()=>{let e=document.documentElement;r===`system`?(e.removeAttribute(`data-theme`),localStorage.removeItem(Ft)):(e.setAttribute(`data-theme`,r),localStorage.setItem(Ft,r))},[r]),(0,_.useEffect)(()=>{let e=!1,t=async()=>{try{let t=await fetch(`${Pt}/healthz`);if(!t.ok)return;let n=zt(await t.json());!e&&n&&o(n)}catch{}};t();let n=setInterval(t,3e4);return()=>{e=!0,clearInterval(n)}},[]);let u=()=>i(e=>e===`light`?`dark`:e===`dark`?`system`:`light`),d=Lt[r],f=a??`2.6.31-preview.20260707`,p=me.find(e=>e.code===s)?.name??`English`,m=()=>{let e=me.map(e=>e.code);c(e[(e.indexOf(s)+1)%e.length])},[h,g]=(0,_.useState)(!1);return(0,S.jsxs)(`div`,{className:`app`,children:[(0,S.jsxs)(`aside`,{className:`sidebar`,children:[(0,S.jsxs)(`div`,{className:`brand`,children:[(0,S.jsx)(`span`,{className:`brand-logo`,role:`img`,"aria-label":`opencodex logo`}),(0,S.jsx)(`span`,{className:`name`,children:`opencodex`}),(0,S.jsxs)(`span`,{className:`ver`,children:[`v`,f]})]}),(0,S.jsx)(`nav`,{children:It.map(({id:t,tkey:r,Icon:i})=>(0,S.jsxs)(`button`,{className:`nav-item${e===t?` active`:``}`,"data-page":t,onClick:()=>n(t),"aria-current":e===t?`page`:void 0,children:[(0,S.jsx)(i,{}),` `,l(r)]},t))}),(0,S.jsxs)(`div`,{className:`sidebar-foot`,children:[(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle`,onClick:m,"aria-label":`${l(`lang.label`)}: ${p}`,title:`${l(`lang.label`)}: ${p}`,children:[(0,S.jsx)(de,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:p})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle`,onClick:u,"aria-label":`${l(`theme.label`)}: ${l(Rt[r])}`,title:`${l(`theme.label`)}: ${l(Rt[r])}`,children:[(0,S.jsx)(d,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(Rt[r])})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle stop-toggle`,onClick:async()=>{if(confirm(l(`dash.stopConfirm`))){g(!0);try{await fetch(`${Pt}/api/stop`,{method:`POST`})}catch{}}},disabled:h,"aria-label":l(`dash.stop`),title:l(`dash.stop`),children:[(0,S.jsx)(F,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(h?`dash.stopping`:`dash.stop`)})]}),(0,S.jsxs)(`a`,{className:`sidebar-link`,href:`https://github.com/lidge-jun/opencodex`,target:`_blank`,rel:`noreferrer`,children:[(0,S.jsx)(P,{}),` `,l(`common.github`)]})]})]}),(0,S.jsx)(`main`,{className:`main`,children:(0,S.jsxs)(`div`,{className:`main-inner`,children:[e===`dashboard`&&(0,S.jsx)(Oe,{apiBase:Pt}),e===`providers`&&(0,S.jsx)(We,{apiBase:Pt}),e===`models`&&(0,S.jsx)(Ye,{apiBase:Pt}),e===`subagents`&&(0,S.jsx)(Xe,{apiBase:Pt}),e===`logs`&&(0,S.jsx)(ot,{apiBase:Pt}),e===`usage`&&(0,S.jsx)(pt,{apiBase:Pt}),e===`codex-auth`&&(0,S.jsx)(ht,{apiBase:Pt}),e===`api`&&(0,S.jsx)(bt,{apiBase:Pt})]})})]})}v.createRoot(document.getElementById(`root`)).render((0,S.jsx)(_.StrictMode,{children:(0,S.jsx)(ye,{children:(0,S.jsx)(Vt,{})})}));
+  }'`})]})]})}var xt=`opencodex-api-token`,St=!1,Ct=null;function wt(e){try{let t=e instanceof Request?e.url:String(e);return new URL(t,window.location.href).pathname}catch{return null}}function Tt(e){let t=wt(e);return!!t&&(t.startsWith(`/api/`)||t.startsWith(`/v1/`))}function Et(){try{return sessionStorage.getItem(xt)?.trim()||null}catch{return null}}function Dt(e){try{sessionStorage.setItem(xt,e)}catch{}}function Ot(){try{sessionStorage.removeItem(xt)}catch{}}function kt(e,t,n){let r=new Headers(t?.headers??(e instanceof Request?e.headers:void 0));return r.set(`X-OpenCodex-API-Key`,n),e instanceof Request?[new Request(e,{headers:r}),t?{...t,headers:r}:void 0]:[e,{...t,headers:r}]}async function At(){return Ct||(Ct=Promise.resolve().then(()=>window.prompt(`OpenCodex API token`)?.trim()||null).finally(()=>{Ct=null}),Ct)}function jt(){if(St)return;St=!0;let e=window.fetch.bind(window);window.fetch=async(t,n)=>{if(!Tt(t))return e(t,n);let r=Et(),[i,a]=r?kt(t,n,r):[t,n],o=await e(i,a);if(o.status!==401)return o;r&&Ot();let s=await At();if(!s)return o;Dt(s);let[c,l]=kt(t,n,s),u=await e(c,l);return u.status===401&&Ot(),u}}jt();var Mt=new Set([`dashboard`,`providers`,`models`,`subagents`,`logs`,`usage`,`codex-auth`,`api`]);function Nt(){let e=location.hash.replace(/^#\/?/,``);return Mt.has(e)?e:`dashboard`}var Pt=``,Ft=`ocx-theme`,It=[{id:`dashboard`,tkey:`nav.dashboard`,Icon:w},{id:`providers`,tkey:`nav.providers`,Icon:T},{id:`models`,tkey:`nav.models`,Icon:E},{id:`subagents`,tkey:`nav.subagents`,Icon:ee},{id:`logs`,tkey:`nav.logs`,Icon:D},{id:`usage`,tkey:`nav.usage`,Icon:te},{id:`codex-auth`,tkey:`nav.codexAuth`,Icon:L},{id:`api`,tkey:`nav.api`,Icon:de}],Lt={light:ce,dark:le,system:ue},Rt={light:`theme.light`,dark:`theme.dark`,system:`theme.system`};function zt(e){if(!e||typeof e!=`object`||!(`version`in e))return null;let t=e.version;return typeof t==`string`&&t.length>0?t:null}function Bt(){let e=localStorage.getItem(Ft);return e===`light`||e===`dark`?e:`system`}function Vt(){let[e,t]=(0,_.useState)(Nt),n=e=>{location.hash=e,t(e)},[r,i]=(0,_.useState)(Bt),[a,o]=(0,_.useState)(null),{locale:s,setLocale:c}=be(),l=xe();(0,_.useEffect)(()=>{let e=()=>t(Nt());return window.addEventListener(`hashchange`,e),()=>window.removeEventListener(`hashchange`,e)},[]),(0,_.useEffect)(()=>{let e=document.documentElement;r===`system`?(e.removeAttribute(`data-theme`),localStorage.removeItem(Ft)):(e.setAttribute(`data-theme`,r),localStorage.setItem(Ft,r))},[r]),(0,_.useEffect)(()=>{let e=!1,t=async()=>{try{let t=await fetch(`${Pt}/healthz`);if(!t.ok)return;let n=zt(await t.json());!e&&n&&o(n)}catch{}};t();let n=setInterval(t,3e4);return()=>{e=!0,clearInterval(n)}},[]);let u=()=>i(e=>e===`light`?`dark`:e===`dark`?`system`:`light`),d=Lt[r],f=a??`2.6.32`,p=me.find(e=>e.code===s)?.name??`English`,m=()=>{let e=me.map(e=>e.code);c(e[(e.indexOf(s)+1)%e.length])},[h,g]=(0,_.useState)(!1);return(0,S.jsxs)(`div`,{className:`app`,children:[(0,S.jsxs)(`aside`,{className:`sidebar`,children:[(0,S.jsxs)(`div`,{className:`brand`,children:[(0,S.jsx)(`span`,{className:`brand-logo`,role:`img`,"aria-label":`opencodex logo`}),(0,S.jsx)(`span`,{className:`name`,children:`opencodex`}),(0,S.jsxs)(`span`,{className:`ver`,children:[`v`,f]})]}),(0,S.jsx)(`nav`,{children:It.map(({id:t,tkey:r,Icon:i})=>(0,S.jsxs)(`button`,{className:`nav-item${e===t?` active`:``}`,"data-page":t,onClick:()=>n(t),"aria-current":e===t?`page`:void 0,children:[(0,S.jsx)(i,{}),` `,l(r)]},t))}),(0,S.jsxs)(`div`,{className:`sidebar-foot`,children:[(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle`,onClick:m,"aria-label":`${l(`lang.label`)}: ${p}`,title:`${l(`lang.label`)}: ${p}`,children:[(0,S.jsx)(de,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:p})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle`,onClick:u,"aria-label":`${l(`theme.label`)}: ${l(Rt[r])}`,title:`${l(`theme.label`)}: ${l(Rt[r])}`,children:[(0,S.jsx)(d,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(Rt[r])})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle stop-toggle`,onClick:async()=>{if(confirm(l(`dash.stopConfirm`))){g(!0);try{await fetch(`${Pt}/api/stop`,{method:`POST`})}catch{}}},disabled:h,"aria-label":l(`dash.stop`),title:l(`dash.stop`),children:[(0,S.jsx)(F,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(h?`dash.stopping`:`dash.stop`)})]}),(0,S.jsxs)(`a`,{className:`sidebar-link`,href:`https://github.com/lidge-jun/opencodex`,target:`_blank`,rel:`noreferrer`,children:[(0,S.jsx)(P,{}),` `,l(`common.github`)]})]})]}),(0,S.jsx)(`main`,{className:`main`,children:(0,S.jsxs)(`div`,{className:`main-inner`,children:[e===`dashboard`&&(0,S.jsx)(Oe,{apiBase:Pt}),e===`providers`&&(0,S.jsx)(We,{apiBase:Pt}),e===`models`&&(0,S.jsx)(Ye,{apiBase:Pt}),e===`subagents`&&(0,S.jsx)(Xe,{apiBase:Pt}),e===`logs`&&(0,S.jsx)(ot,{apiBase:Pt}),e===`usage`&&(0,S.jsx)(pt,{apiBase:Pt}),e===`codex-auth`&&(0,S.jsx)(ht,{apiBase:Pt}),e===`api`&&(0,S.jsx)(bt,{apiBase:Pt})]})})]})}v.createRoot(document.getElementById(`root`)).render((0,S.jsx)(_.StrictMode,{children:(0,S.jsx)(ye,{children:(0,S.jsx)(Vt,{})})}));

package/gui/dist/index.html

@@ -16,7 +16,7 @@
         } catch (e) {}
       })();
     </script>
-    <script type="module" crossorigin src="/assets/index-CWujz83O.js"></script>
+    <script type="module" crossorigin src="/assets/index-D_JZzI0r.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-ByGC8-Bm.css">
   </head>
   <body>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bitkyc08/opencodex",
-  "version": "2.6.31-preview.20260707",
+  "version": "2.6.32",
   "description": "Universal provider proxy for OpenAI Codex — use any LLM with Codex CLI/App/SDK",
   "type": "module",
   "main": "./bin/package-main.mjs",

package/src/adapters/kiro.ts

@@ -180,7 +180,7 @@ function kiroThinkingBudget(parsed: OcxParsedRequest): number | undefined {
     low: 0.20,
     medium: 0.50,
     high: 0.80,
-    xhigh: 0.95,
+    xhigh: 0.90,
     max: 0.95,
   };
   const ratio = percent[effort];

package/src/bridge.ts

@@ -20,8 +20,15 @@ function responsesUsage(usage: OcxUsage | undefined): Record<string, unknown> {
     output_tokens: usage.outputTokens,
     total_tokens: usageDisplayTotalTokens(usage) ?? inputTokens + usage.outputTokens,
   };
+  const inputDetails: Record<string, number> = {};
   if (usage.cachedInputTokens !== undefined) {
-    out.input_tokens_details = { cached_tokens: usage.cachedInputTokens };
+    inputDetails.cached_tokens = usage.cachedInputTokens;
+  }
+  if (usage.cacheCreationInputTokens !== undefined) {
+    inputDetails.cache_write_tokens = usage.cacheCreationInputTokens;
+  }
+  if (Object.keys(inputDetails).length > 0) {
+    out.input_tokens_details = inputDetails;
   }
   if (usage.reasoningOutputTokens !== undefined) {
     out.output_tokens_details = { reasoning_tokens: usage.reasoningOutputTokens };

package/src/codex/account-store.ts

@@ -44,7 +44,10 @@ function isCredentialRecord(value: unknown): value is CodexAccountCredentialReco
     && (value.credential === undefined || isCredential(value.credential))
     && (value.refreshGrantFingerprint === undefined || typeof value.refreshGrantFingerprint === "string")
     && (value.deletedAt === undefined || typeof value.deletedAt === "number")
-    && (value.replacedAt === undefined || typeof value.replacedAt === "number");
+    && (value.replacedAt === undefined || typeof value.replacedAt === "number")
+    && (value.lastCodexValidatedAt === undefined || typeof value.lastCodexValidatedAt === "number")
+    && (value.lastCodexValidationStatus === undefined || value.lastCodexValidationStatus === "ok" || value.lastCodexValidationStatus === "failed")
+    && (value.lastCodexValidationError === undefined || typeof value.lastCodexValidationError === "string");
 }
 
 export function refreshGrantFingerprintForToken(refreshToken: string): string {
@@ -98,6 +101,17 @@ function persist(store: CodexAccountStore): void {
   atomicWriteFile(codexAccountsPath(), JSON.stringify(store, null, 2) + "\n");
 }
 
+function preservedValidationMetadata(record: CodexAccountCredentialRecord | undefined): Pick<
+  CodexAccountCredentialRecord,
+  "lastCodexValidatedAt" | "lastCodexValidationStatus" | "lastCodexValidationError"
+> {
+  return {
+    ...(record?.lastCodexValidatedAt !== undefined ? { lastCodexValidatedAt: record.lastCodexValidatedAt } : {}),
+    ...(record?.lastCodexValidationStatus !== undefined ? { lastCodexValidationStatus: record.lastCodexValidationStatus } : {}),
+    ...(record?.lastCodexValidationError !== undefined ? { lastCodexValidationError: record.lastCodexValidationError } : {}),
+  };
+}
+
 export function getCodexAccountCredential(id: string): CodexAccountCredentials | null {
   const record = readCodexAccountRecord(id);
   if (!record || record.deletedAt != null) return null;
@@ -115,6 +129,32 @@ export function saveCodexAccountCredential(id: string, cred: CodexAccountCredent
     generation: (current?.generation ?? 0) + 1,
     refreshGrantFingerprint,
     replacedAt: current ? Date.now() : undefined,
+    ...preservedValidationMetadata(current),
+  };
+  persist(store);
+}
+
+export function markCodexAccountValidated(id: string, atMs: number = Date.now()): void {
+  const store = loadCodexAccountRecordStore();
+  const current = store[id];
+  if (!current || current.deletedAt != null || !current.credential) return;
+  store[id] = {
+    ...current,
+    lastCodexValidatedAt: atMs,
+    lastCodexValidationStatus: "ok",
+    lastCodexValidationError: undefined,
+  };
+  persist(store);
+}
+
+export function markCodexAccountValidationFailed(id: string, reason: string): void {
+  const store = loadCodexAccountRecordStore();
+  const current = store[id];
+  if (!current || current.deletedAt != null || !current.credential) return;
+  store[id] = {
+    ...current,
+    lastCodexValidationStatus: "failed",
+    lastCodexValidationError: reason,
   };
   persist(store);
 }
@@ -154,6 +194,7 @@ export function saveCodexAccountCredentialIfGeneration(
     generation: generation + 1,
     refreshGrantFingerprint,
     replacedAt: current.replacedAt,
+    ...preservedValidationMetadata(current),
   };
   persist(store);
   return true;

package/src/codex/auth-api.ts

@@ -3,6 +3,7 @@ import { withCodexAccountLogLabel } from "./account-label";
 import {
   getCodexAccountCredential,
   getValidCodexToken,
+  markCodexAccountValidated,
   saveCodexAccountCredential,
   CodexCredentialGenerationConflictError,
   CodexCredentialRefreshLockTimeoutError,
@@ -26,6 +27,7 @@ export { clearAccountQuota, getAccountQuota, parseUsageQuota, updateAccountQuota
 import { extractAccountId, decodeJwtPayload } from "../oauth/chatgpt";
 import { MAIN_CODEX_ACCOUNT_ID, setMainAccountPlan } from "./main-account";
 import { maskEmail } from "../lib/privacy";
+import { codexWarmupFailureReason, warmCodexAccount } from "./warmup";
 export { maskEmail } from "../lib/privacy";
 import type { CodexAccount, OcxConfig } from "../types";
 
@@ -138,6 +140,27 @@ function manualImportDisabledResponse(): Response {
   }, 403);
 }
 
+async function verifyCodexAccountWarmup(
+  accountId: string,
+  accessToken: string,
+  chatgptAccountId: string,
+): Promise<{ ok: true; validatedAt: number } | { ok: false; response: Response }> {
+  try {
+    await warmCodexAccount({ accessToken, chatgptAccountId });
+    return { ok: true, validatedAt: Date.now() };
+  } catch (err) {
+    return {
+      ok: false,
+      response: jsonResponse({
+        error: "Codex account warmup failed. Reauthenticate the account and try again.",
+        code: "codex_warmup_failed",
+        reason: codexWarmupFailureReason(err),
+        accountId,
+      }, 401),
+    };
+  }
+}
+
 function expireCodexAuthFlow(flowId: string | null, error = "Login cancelled"): void {
   if (!flowId) return;
   codexAuthLoginState.set(flowId, { status: "error", error, doneAt: Date.now() });
@@ -385,12 +408,15 @@ export async function handleCodexAuthAPI(
     // 4.2: use JWT exp for expiresAt instead of hardcoded 1 hour
     const payload = decodeJwtPayload(body.accessToken);
     const exp = typeof payload?.exp === "number" ? payload.exp * 1000 : Date.now() + 3600_000;
+    const warmup = await verifyCodexAccountWarmup(body.id, body.accessToken, derivedAccountId);
+    if (!warmup.ok) return warmup.response;
     saveCodexAccountCredential(body.id, {
       accessToken: body.accessToken,
       refreshToken: body.refreshToken,
       expiresAt: exp,
       chatgptAccountId: derivedAccountId,
     });
+    markCodexAccountValidated(body.id, warmup.validatedAt);
     clearAccountNeedsReauth(body.id);
     accounts.push(withCodexAccountLogLabel({ id: body.id, email: body.email, plan: body.plan, isMain: false }, accounts));
     runtimeConfig.codexAccounts = accounts;
@@ -590,12 +616,25 @@ export async function handleCodexAuthAPI(
                 break;
               }
 
+              const warmup = await verifyCodexAccountWarmup(accountId, cred.access, oauthAccountId);
+              if (!warmup.ok) {
+                const body = await warmup.response.json().catch(() => ({})) as { error?: string; reason?: string };
+                codexAuthLoginState.set(flowId, {
+                  status: "error",
+                  error: body.reason ? `${body.error ?? "Codex account warmup failed"} (${body.reason})` : body.error ?? "Codex account warmup failed",
+                  doneAt: Date.now(),
+                });
+                completed = true;
+                break;
+              }
+
               saveCodexAccountCredential(accountId, {
                 accessToken: cred.access,
                 refreshToken: cred.refresh,
                 expiresAt: cred.expires,
                 chatgptAccountId: oauthAccountId,
               });
+              markCodexAccountValidated(accountId, warmup.validatedAt);
               clearAccountNeedsReauth(accountId);
               if (quota) {
                 updateAccountQuota(

package/src/codex/catalog.ts

@@ -78,9 +78,13 @@ function isDefaultCatalogPath(path: string): boolean {
  */
 export const NATIVE_OPENAI_MODELS = [
   "gpt-5.5", "gpt-5.4", "gpt-5.4-mini", "gpt-5.3-codex-spark",
+  "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna",
 ];
 
-const DOCUMENTED_NATIVE_OPENAI_ADDITIONS = ["gpt-5.3-codex-spark"];
+const DOCUMENTED_NATIVE_OPENAI_ADDITIONS = [
+  "gpt-5.3-codex-spark",
+  "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna",
+];
 
 /**
  * The ONLY native OpenAI/Codex slugs opencodex advertises. A user's installed Codex ships extra
@@ -102,10 +106,15 @@ function isUnsupportedOpenAiNativeSlug(slug: string): boolean {
   return /^(?:gpt|codex)-/.test(slug);
 }
 
+const NATIVE_GPT56_CONTEXT_WINDOW = 372_000;
+
 const NATIVE_OPENAI_CONTEXT_OVERRIDES: Record<string, { contextWindow?: number; maxContextWindow?: number }> = {
   "gpt-5.5": { contextWindow: 272_000, maxContextWindow: 272_000 },
   "gpt-5.4": { contextWindow: 1_000_000, maxContextWindow: 1_000_000 },
   "gpt-5.3-codex-spark": { contextWindow: 128_000, maxContextWindow: 128_000 },
+  "gpt-5.6-sol": { contextWindow: NATIVE_GPT56_CONTEXT_WINDOW, maxContextWindow: NATIVE_GPT56_CONTEXT_WINDOW },
+  "gpt-5.6-terra": { contextWindow: NATIVE_GPT56_CONTEXT_WINDOW, maxContextWindow: NATIVE_GPT56_CONTEXT_WINDOW },
+  "gpt-5.6-luna": { contextWindow: NATIVE_GPT56_CONTEXT_WINDOW, maxContextWindow: NATIVE_GPT56_CONTEXT_WINDOW },
 };
 
 /**
@@ -466,8 +475,8 @@ export function loadCatalogTemplate(): RawEntry | null {
 }
 
 /**
- * Codex only accepts its native labels in the catalog. Provider-specific wire values (e.g. Z.AI
- * `max`) are mapped at request time by src/reasoning-effort.ts, never advertised directly here.
+ * Codex accepts its native labels plus model-defined effort strings such as `max` in current builds.
+ * Provider-specific aliases still map at request time by src/reasoning-effort.ts.
  */
 const ROUTED_REASONING_LEVELS = CODEX_REASONING_LEVELS;
 
@@ -501,6 +510,20 @@ function applyReasoningLevels(entry: RawEntry, effortsOverride?: string[]): void
   entry.default_reasoning_level = efforts.includes("medium") ? "medium" : efforts.includes("high") ? "high" : efforts[0];
 }
 
+function isGpt56NativeSlug(slug: string): boolean {
+  return !slug.includes("/") && slug.startsWith("gpt-5.6-");
+}
+
+function ensureMaxReasoningLevel(entry: RawEntry): void {
+  const levels = Array.isArray(entry.supported_reasoning_levels)
+    ? entry.supported_reasoning_levels as Array<{ effort?: string }>
+    : [];
+  if (levels.some(level => level.effort === "max")) return;
+  const maxLevel = ROUTED_REASONING_LEVELS.find(level => level.effort === "max")
+    ?? { effort: "max", description: "Maximum reasoning for the hardest problems" };
+  entry.supported_reasoning_levels = [...levels, maxLevel];
+}
+
 function deriveEntry(template: RawEntry | null, slug: string, desc: string, priority: number, model?: CatalogModel): RawEntry {
   if (template) {
     const e = JSON.parse(JSON.stringify(template)) as RawEntry;
@@ -512,7 +535,7 @@ function deriveEntry(template: RawEntry | null, slug: string, desc: string, prio
     if ("upgrade" in e) e.upgrade = null;
     delete e.availability_nux; // don't replay another model's "now available" NUX
     // Routed (namespaced) models inherit the gpt template — correct its OpenAI/GPT identity
-    // and advertise the reasoning ladder Codex accepts (low/medium/high/xhigh).
+    // and advertise the reasoning ladder Codex accepts.
     if (slug.includes("/")) {
       const modelName = slug.slice(slug.indexOf("/") + 1);
       if (typeof e.base_instructions === "string") {
@@ -529,6 +552,7 @@ function deriveEntry(template: RawEntry | null, slug: string, desc: string, prio
       applyCatalogModelMetadata(e, model);
     } else {
       applyNativeOpenAiContextOverride(e);
+      if (isGpt56NativeSlug(slug)) ensureMaxReasoningLevel(e);
     }
     return ensureStrictCatalogFields(normalizeServiceTiers(e));
   }
@@ -540,7 +564,7 @@ function deriveEntry(template: RawEntry | null, slug: string, desc: string, prio
     ...(slug.includes("/") ? { web_search_tool_type: "text_and_image", supports_search_tool: true } : {}),
   };
   if (slug.includes("/")) applyReasoningLevels(entry, model?.reasoningEfforts);
-  else applyReasoningLevels(entry);
+  else applyReasoningLevels(entry, isGpt56NativeSlug(slug) ? undefined : ["low", "medium", "high", "xhigh"]);
   applyJawcodeCatalogMetadata(entry, slug, model?.contextCap);
   applyCatalogModelMetadata(entry, model);
   applyNativeOpenAiContextOverride(entry);
@@ -744,8 +768,8 @@ function catalogHintsFromModelsApiItem(providerName: string, item: ProviderModel
   const reasoningEfforts = capabilities && typeof capabilities.reasoning_effort === "boolean"
     ? (capabilities.reasoning_effort
       ? ((providerName === "neuralwatt" || providerName === "zai") && isGlm52ModelId(item.id)
-        ? ["low", "medium", "high", "xhigh"]
-        : ["low", "medium", "high"])
+        ? ["low", "medium", "high", "xhigh", "max"]
+        : ["low", "medium", "high", "xhigh", "max"])
       : [])
     : undefined;
   const inputModalities = capabilities && typeof capabilities.vision === "boolean"

package/src/codex/warmup.ts

@@ -0,0 +1,137 @@
+export class CodexWarmupError extends Error {
+  code: "http_status" | "missing_body" | "stream_failed" | "stream_incomplete" | "stream_error" | "invalid_sse" | "no_terminal" | "transport";
+  status?: number;
+
+  constructor(
+    code: CodexWarmupError["code"],
+    message = "Codex warmup failed",
+    options: { status?: number; cause?: unknown } = {},
+  ) {
+    super(message);
+    this.name = "CodexWarmupError";
+    this.code = code;
+    this.status = options.status;
+    if (options.cause !== undefined) this.cause = options.cause;
+  }
+}
+
+export interface CodexWarmupOptions {
+  accessToken: string;
+  chatgptAccountId: string;
+  model?: string;
+  timeoutMs?: number;
+}
+
+const CODEX_RESPONSES_URL = "https://chatgpt.com/backend-api/codex/responses";
+const DEFAULT_MODEL = "gpt-5.4-mini";
+const DEFAULT_TIMEOUT_MS = 30_000;
+
+function safeWarmupReason(err: unknown): string {
+  if (err instanceof CodexWarmupError) {
+    return err.status ? `${err.code}:${err.status}` : err.code;
+  }
+  return "transport";
+}
+
+export function codexWarmupFailureReason(err: unknown): string {
+  return safeWarmupReason(err);
+}
+
+function eventTypeFromData(data: unknown): string | undefined {
+  if (!data || typeof data !== "object") return undefined;
+  const record = data as Record<string, unknown>;
+  return typeof record.type === "string" ? record.type : undefined;
+}
+
+function parseSseFrame(frame: string): unknown | null {
+  const dataLines = frame
+    .split(/\r?\n/)
+    .filter(line => line.startsWith("data:"))
+    .map(line => line.slice(5).trimStart());
+  if (dataLines.length === 0) return null;
+  const data = dataLines.join("\n").trim();
+  if (!data || data === "[DONE]") return null;
+  try {
+    return JSON.parse(data) as unknown;
+  } catch (err) {
+    throw new CodexWarmupError("invalid_sse", "Codex warmup received invalid SSE", { cause: err });
+  }
+}
+
+async function drainWarmupSse(body: ReadableStream<Uint8Array>): Promise<void> {
+  const reader = body.getReader();
+  const decoder = new TextDecoder();
+  let buffer = "";
+
+  try {
+    for (;;) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+
+      for (;;) {
+        const frameEnd = buffer.search(/\r?\n\r?\n/);
+        if (frameEnd < 0) break;
+        const frame = buffer.slice(0, frameEnd);
+        const delimiterLength = buffer[frameEnd] === "\r" ? 4 : 2;
+        buffer = buffer.slice(frameEnd + delimiterLength);
+        const parsed = parseSseFrame(frame);
+        const type = eventTypeFromData(parsed);
+        if (type === "response.completed") return;
+        if (type === "response.failed") throw new CodexWarmupError("stream_failed");
+        if (type === "response.incomplete") throw new CodexWarmupError("stream_incomplete");
+        if (type === "error") throw new CodexWarmupError("stream_error");
+      }
+    }
+
+    if (buffer.trim()) {
+      const parsed = parseSseFrame(buffer);
+      const type = eventTypeFromData(parsed);
+      if (type === "response.completed") return;
+      if (type === "response.failed") throw new CodexWarmupError("stream_failed");
+      if (type === "response.incomplete") throw new CodexWarmupError("stream_incomplete");
+      if (type === "error") throw new CodexWarmupError("stream_error");
+    }
+
+    throw new CodexWarmupError("no_terminal", "Codex warmup ended before completion");
+  } finally {
+    reader.releaseLock();
+  }
+}
+
+export async function warmCodexAccount(options: CodexWarmupOptions): Promise<void> {
+  let res: Response;
+  try {
+    res = await fetch(CODEX_RESPONSES_URL, {
+      method: "POST",
+      headers: {
+        Authorization: `Bearer ${options.accessToken}`,
+        "ChatGPT-Account-Id": options.chatgptAccountId,
+        "Content-Type": "application/json",
+      },
+      body: JSON.stringify({
+        model: options.model?.trim() || DEFAULT_MODEL,
+        instructions: "Reply with OK.",
+        input: "hi",
+        stream: true,
+        store: false,
+      }),
+      signal: AbortSignal.timeout(options.timeoutMs ?? DEFAULT_TIMEOUT_MS),
+    });
+  } catch (err) {
+    throw new CodexWarmupError("transport", "Codex warmup request failed", { cause: err });
+  }
+
+  if (!res.ok) {
+    await res.body?.cancel().catch(() => {});
+    throw new CodexWarmupError("http_status", "Codex warmup was rejected", { status: res.status });
+  }
+  if (!res.body) throw new CodexWarmupError("missing_body");
+
+  try {
+    await drainWarmupSse(res.body);
+  } finally {
+    await res.body?.cancel().catch(() => {});
+  }
+}
+

package/src/oauth/token-guardian.ts

@@ -18,9 +18,12 @@ import { getValidAccessTokenForAccount, listOAuthProviders, OAuthLoginRequiredEr
 import {
   getValidCodexToken,
   listCodexAccountIds,
+  markCodexAccountValidated,
+  markCodexAccountValidationFailed,
   readCodexAccountRecord,
   TokenRefreshError,
 } from "../codex/account-store";
+import { codexWarmupFailureReason, warmCodexAccount } from "../codex/warmup";
 
 export interface TokenGuardianHandle {
   stop(): void;
@@ -29,6 +32,7 @@ export interface TokenGuardianHandle {
 export interface GuardianSweepResult {
   enabled: boolean;
   refreshed: string[];
+  warmed: string[];
   failed: string[];
   skippedBackoff: string[];
 }
@@ -40,6 +44,8 @@ const DEFAULTS = {
   leadSeconds: 900,
   failureBackoffBaseSeconds: 300,
   failureBackoffMaxSeconds: 3600,
+  codexWarmupMaxAgeSeconds: 691_200, // 8d — matches Codex managed-auth last_refresh cadence.
+  codexWarmupModel: "gpt-5.4-mini",
 };
 
 interface BackoffEntry {
@@ -67,6 +73,9 @@ function resolved(g: OcxTokenGuardianConfig | undefined) {
     leadSeconds: num(g?.leadSeconds, DEFAULTS.leadSeconds, 0),
     backoffBaseSeconds: num(g?.failureBackoffBaseSeconds, DEFAULTS.failureBackoffBaseSeconds, 0),
     backoffMaxSeconds: num(g?.failureBackoffMaxSeconds, DEFAULTS.failureBackoffMaxSeconds, 0),
+    codexWarmupEnabled: g?.codexWarmupEnabled === true,
+    codexWarmupMaxAgeSeconds: num(g?.codexWarmupMaxAgeSeconds, DEFAULTS.codexWarmupMaxAgeSeconds, 60),
+    codexWarmupModel: g?.codexWarmupModel?.trim() || DEFAULTS.codexWarmupModel,
   };
 }
 
@@ -106,7 +115,7 @@ async function runWithConcurrency(tasks: Array<() => Promise<void>>, limit: numb
 export async function guardianSweep(nowMs: number = Date.now()): Promise<GuardianSweepResult> {
   const config: OcxConfig = loadConfig();
   const g = config.tokenGuardian;
-  const result: GuardianSweepResult = { enabled: !!g?.enabled, refreshed: [], failed: [], skippedBackoff: [] };
+  const result: GuardianSweepResult = { enabled: !!g?.enabled, refreshed: [], warmed: [], failed: [], skippedBackoff: [] };
   if (!g?.enabled) return result;
 
   const opts = resolved(g);
@@ -142,18 +151,34 @@ export async function guardianSweep(nowMs: number = Date.now()): Promise<Guardia
   if (resolveRefreshPolicy("chatgpt", config) === "proactive") {
     for (const id of listCodexAccountIds()) {
       const record = readCodexAccountRecord(id);
-      const cred = record?.deletedAt == null ? record?.credential : undefined;
+      if (!record || record.deletedAt != null) continue;
+      const cred = record.credential;
       if (!cred) continue;
-      if (cred.expiresAt > nowMs + horizonMs) continue;
+      const needsRefresh = cred.expiresAt <= nowMs + horizonMs;
+      const needsWarmup = opts.codexWarmupEnabled
+        && (record.lastCodexValidatedAt === undefined || nowMs - record.lastCodexValidatedAt > opts.codexWarmupMaxAgeSeconds * 1000);
+      if (!needsRefresh && !needsWarmup) continue;
       const key = `codex:${id}`;
       if (inBackoff(key, nowMs)) { result.skippedBackoff.push(key); continue; }
       tasks.push(async () => {
         try {
-          await getValidCodexToken(id);
+          const token = await getValidCodexToken(id);
+          if (needsRefresh) result.refreshed.push(key);
+          if (needsWarmup) {
+            await warmCodexAccount({
+              accessToken: token.accessToken,
+              chatgptAccountId: token.chatgptAccountId,
+              model: opts.codexWarmupModel,
+            });
+            markCodexAccountValidated(id, Date.now());
+            result.warmed.push(key);
+          }
           backoff.delete(key);
-          result.refreshed.push(key);
         } catch (err) {
           const permanent = err instanceof TokenRefreshError && (err.reason === "revoked" || err.reason === "expired");
+          if (needsWarmup && !(err instanceof TokenRefreshError)) {
+            markCodexAccountValidationFailed(id, codexWarmupFailureReason(err));
+          }
           recordFailure(key, nowMs, opts.backoffBaseSeconds, opts.backoffMaxSeconds, permanent);
           result.failed.push(key);
         }
@@ -186,8 +211,8 @@ export function startTokenGuardian(): TokenGuardianHandle {
   const runSweep = () => {
     void guardianSweep()
       .then(r => {
-        if (r.enabled && (r.refreshed.length || r.failed.length)) {
-          console.log(`🛡️  token-guardian: refreshed ${r.refreshed.length}, failed ${r.failed.length}`);
+        if (r.enabled && (r.refreshed.length || r.warmed.length || r.failed.length)) {
+          console.log(`🛡️  token-guardian: refreshed ${r.refreshed.length}, warmed ${r.warmed.length}, failed ${r.failed.length}`);
         }
       })
       .catch(err => console.log(`token-guardian sweep error: ${err instanceof Error ? err.message : String(err)}`))

package/src/providers/kiro-models.ts

@@ -38,10 +38,10 @@ export const KIRO_MODEL_CONTEXT_WINDOWS: Record<string, number> = {
   "qwen3-coder-next": 256_000,
 };
 
-const KIRO_REASONING_EFFORTS = ["low", "medium", "high", "xhigh"];
+const KIRO_REASONING_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
 
-// Codex does not accept raw "max" in catalog metadata; Kiro xhigh already maps to the maximum
-// fake-thinking budget in src/adapters/kiro.ts.
+// Kiro has no upstream reasoning_effort enum; these labels map to fake-thinking budgets in
+// src/adapters/kiro.ts.
 export const KIRO_MODEL_REASONING_EFFORTS: Record<string, string[]> = Object.fromEntries(
   KIRO_MODELS.map(id => [id, KIRO_REASONING_EFFORTS]),
 );

package/src/providers/registry.ts

@@ -60,24 +60,25 @@ export type ProviderConfigSeed = Pick<
   | "googleMode" | "project" | "location"
 >;
 
-
-const OLLAMA_REASONING_MAP: Record<string, string> = { xhigh: "max" };
-
 // Shared between the OAuth (Claude account) and API-key Anthropic entries so both expose the
 // same static model seed.
 const ANTHROPIC_MODELS = ["claude-sonnet-5", "claude-opus-4-8", "claude-opus-4-7", "claude-opus-4-6", "claude-sonnet-4-6", "claude-haiku-4-5"];
 const ANTHROPIC_MODEL_CONTEXT_WINDOWS: Record<string, number> = { "claude-sonnet-5": 1_000_000 };
 
 const ZAI_GLM_52_MODELS = ["glm-5.2", "glm-5.2[1m]"];
-const ZAI_GLM_52_REASONING_EFFORTS = ["low", "medium", "high", "xhigh"];
-const ZAI_GLM_52_REASONING_MAP: Record<string, string> = {
-  none: "none",
-  minimal: "none",
-  low: "high",
-  medium: "high",
-  high: "high",
-  xhigh: "max",
-  max: "max",
+const ZAI_GLM_52_REASONING_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
+const OPENAI_GPT56_MODELS = ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"];
+const OPENAI_GPT56_CONTEXT_WINDOW = 372_000;
+const OPENAI_GPT56_CONTEXT_WINDOWS = {
+  "gpt-5.6-sol": OPENAI_GPT56_CONTEXT_WINDOW,
+  "gpt-5.6-terra": OPENAI_GPT56_CONTEXT_WINDOW,
+  "gpt-5.6-luna": OPENAI_GPT56_CONTEXT_WINDOW,
+};
+const OPENROUTER_GPT56_MODELS = OPENAI_GPT56_MODELS.map(id => `openai/${id}`);
+const OPENROUTER_GPT56_CONTEXT_WINDOWS = {
+  "openai/gpt-5.6-sol": OPENAI_GPT56_CONTEXT_WINDOW,
+  "openai/gpt-5.6-terra": OPENAI_GPT56_CONTEXT_WINDOW,
+  "openai/gpt-5.6-luna": OPENAI_GPT56_CONTEXT_WINDOW,
 };
 
 /**
@@ -124,17 +125,8 @@ const UMANS_MODELS = [
   "umans-glm-5.1",
   "umans-qwen3.6-35b-a3b",
 ];
-const UMANS_REASONING_EFFORTS = ["low", "medium", "high", "xhigh"];
-const UMANS_GLM_REASONING_EFFORTS = ["high", "xhigh"];
-const UMANS_GLM_REASONING_MAP: Record<string, string> = {
-  none: "high",
-  minimal: "high",
-  low: "high",
-  medium: "high",
-  high: "high",
-  xhigh: "max",
-  max: "max",
-};
+const UMANS_REASONING_EFFORTS = ["low", "medium", "high", "xhigh", "max"];
+const UMANS_GLM_REASONING_EFFORTS = ["high", "xhigh", "max"];
 const UMANS_TEXT_ONLY_MODELS = ["umans-glm-5.2", "umans-glm-5.1"];
 const UMANS_MODEL_CONTEXT_WINDOWS: Record<string, number> = {
   "umans-coder": 262_144,
@@ -260,7 +252,7 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     modelContextWindows: KIRO_MODEL_CONTEXT_WINDOWS,
     modelReasoningEfforts: KIRO_MODEL_REASONING_EFFORTS,
   },
-  { id: "openai-apikey", label: "OpenAI (API key)", adapter: "openai-responses", baseUrl: "https://api.openai.com/v1", authKind: "key", featured: true, dashboardUrl: "https://platform.openai.com/api-keys", defaultModel: "gpt-5.5" },
+  { id: "openai-apikey", label: "OpenAI (API key)", adapter: "openai-responses", baseUrl: "https://api.openai.com/v1", authKind: "key", featured: true, dashboardUrl: "https://platform.openai.com/api-keys", defaultModel: "gpt-5.5", models: ["gpt-5.5", ...OPENAI_GPT56_MODELS], modelContextWindows: OPENAI_GPT56_CONTEXT_WINDOWS },
   {
     id: "umans",
     label: "Umans AI Coding Plan",
@@ -278,14 +270,10 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
       "umans-coder": UMANS_REASONING_EFFORTS,
       "umans-kimi-k2.7": UMANS_REASONING_EFFORTS,
       "umans-kimi-k2.6": UMANS_REASONING_EFFORTS,
-      "umans-flash": ["low", "medium", "high"],
+      "umans-flash": UMANS_REASONING_EFFORTS,
       "umans-glm-5.2": UMANS_GLM_REASONING_EFFORTS,
       "umans-glm-5.1": UMANS_GLM_REASONING_EFFORTS,
-      "umans-qwen3.6-35b-a3b": ["low", "medium", "high"],
-    },
-    modelReasoningEffortMap: {
-      "umans-glm-5.2": UMANS_GLM_REASONING_MAP,
-      "umans-glm-5.1": UMANS_GLM_REASONING_MAP,
+      "umans-qwen3.6-35b-a3b": UMANS_REASONING_EFFORTS,
     },
     noVisionModels: UMANS_TEXT_ONLY_MODELS,
     escapeBuiltinToolNames: true,
@@ -300,8 +288,9 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
       "kimi-k2.7-code-highspeed": [],
       ...Object.fromEntries(OPENCODE_GO_THINKING_TOGGLE_MODELS.map(id => [id, THINKING_TOGGLE_EFFORTS])),
     },
+    // glm-5.2 uses identity labels now that `max` is a native Codex level (no alias map);
+    // the thinking-toggle map is a REAL wire alias (effort -> enabled/disabled) and stays.
     modelReasoningEffortMap: {
-      "glm-5.2": ZAI_GLM_52_REASONING_MAP,
       ...Object.fromEntries(OPENCODE_GO_THINKING_TOGGLE_MODELS.map(id => [id, THINKING_TOGGLE_MAP])),
     },
     thinkingToggleModels: OPENCODE_GO_THINKING_TOGGLE_MODELS,
@@ -345,12 +334,11 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
       "kimi-k2.6": [],
       "kimi-k2.6-fast": [],
       "kimi-k2.7-code": [],
-      "qwen3.5-397b": ["low", "medium", "high"],
+      "qwen3.5-397b": ["low", "medium", "high", "xhigh", "max"],
       "qwen3.5-397b-fast": [],
-      "qwen3.6-35b": ["low", "medium", "high"],
+      "qwen3.6-35b": ["low", "medium", "high", "xhigh", "max"],
       "qwen3.6-35b-fast": [],
     },
-    modelReasoningEffortMap: { "glm-5.2": ZAI_GLM_52_REASONING_MAP },
     noReasoningModels: ["glm-5.2-fast", "kimi-k2.5-fast", "kimi-k2.6-fast", "qwen3.5-397b-fast", "qwen3.6-35b-fast"],
     noVisionModels: ["glm-5.2", "glm-5.2-fast", "qwen3.5-397b", "qwen3.5-397b-fast"],
     noTemperatureModels: ["kimi-k2.7-code"],
@@ -359,13 +347,13 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     autoToolChoiceOnlyModels: ["kimi-k2.7-code"],
     preserveReasoningContentModels: NEURALWATT_REASONING_HISTORY_MODELS,
   },
-  { id: "openrouter", label: "OpenRouter", adapter: "openai-chat", baseUrl: "https://openrouter.ai/api/v1", authKind: "key", featured: true, dashboardUrl: "https://openrouter.ai/keys", jawcodeBundle: "openrouter", models: ["anthropic/claude-sonnet-5"], modelContextWindows: { "anthropic/claude-sonnet-5": 1_000_000 } },
+  { id: "openrouter", label: "OpenRouter", adapter: "openai-chat", baseUrl: "https://openrouter.ai/api/v1", authKind: "key", featured: true, dashboardUrl: "https://openrouter.ai/keys", jawcodeBundle: "openrouter", models: ["anthropic/claude-sonnet-5", ...OPENROUTER_GPT56_MODELS], modelContextWindows: { "anthropic/claude-sonnet-5": 1_000_000, ...OPENROUTER_GPT56_CONTEXT_WINDOWS } },
   { id: "groq", label: "Groq", adapter: "openai-chat", baseUrl: "https://api.groq.com/openai/v1", authKind: "key", featured: true, dashboardUrl: "https://console.groq.com/keys" },
   { id: "google", label: "Google Gemini", adapter: "google", baseUrl: "https://generativelanguage.googleapis.com", authKind: "key", featured: true, dashboardUrl: "https://aistudio.google.com/apikey", defaultModel: "gemini-3-pro", jawcodeBundle: "google", extraMetadataAliases: ["gemini"] },
   { id: "google-vertex", label: "Google Vertex AI", adapter: "google", baseUrl: "https://aiplatform.googleapis.com", authKind: "key", dashboardUrl: "https://console.cloud.google.com/vertex-ai", defaultModel: "gemini-3-pro", googleMode: "vertex", jawcodeBundle: "google", extraMetadataAliases: ["gemini-vertex"] },
   { id: "google-antigravity", label: "Google Antigravity", adapter: "google", baseUrl: "https://daily-cloudcode-pa.googleapis.com", authKind: "oauth", dashboardUrl: "https://antigravity.google", models: ANTIGRAVITY_MODELS, defaultModel: "gemini-3.5-flash-low", modelContextWindows: ANTIGRAVITY_MODEL_CONTEXT_WINDOWS, googleMode: "cloud-code-assist", jawcodeBundle: "google", extraMetadataAliases: ["antigravity", "gemini-antigravity"] },
   { id: "azure-openai", label: "Azure OpenAI", adapter: "azure-openai", baseUrl: "https://{resource}.openai.azure.com/openai", authKind: "key", featured: true, dashboardUrl: "https://portal.azure.com" },
-  { id: "ollama", label: "Ollama (local)", adapter: "openai-chat", baseUrl: "http://localhost:11434/v1", authKind: "local", featured: true, note: "Local — key usually blank", reasoningEffortMap: OLLAMA_REASONING_MAP },
+  { id: "ollama", label: "Ollama (local)", adapter: "openai-chat", baseUrl: "http://localhost:11434/v1", authKind: "local", featured: true, note: "Local — key usually blank" },
   { id: "vllm", label: "vLLM (local)", adapter: "openai-chat", baseUrl: "http://localhost:8000/v1", authKind: "local", featured: true, note: "Local — key usually blank" },
   { id: "lm-studio", label: "LM Studio (local)", adapter: "openai-chat", baseUrl: "http://localhost:1234/v1", authKind: "local", featured: true, note: "Local — no key needed" },
   {
@@ -412,7 +400,6 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     models: ["glm-5.2", "glm-5.2[1m]", "glm-5.1", "glm-5", "glm-4.6"],
     noVisionModels: ZAI_GLM_52_MODELS,
     modelReasoningEfforts: Object.fromEntries(ZAI_GLM_52_MODELS.map(id => [id, ZAI_GLM_52_REASONING_EFFORTS])),
-    modelReasoningEffortMap: Object.fromEntries(ZAI_GLM_52_MODELS.map(id => [id, ZAI_GLM_52_REASONING_MAP])),
     preserveReasoningContentModels: ZAI_GLM_52_MODELS,
   },
   { id: "nanogpt", label: "NanoGPT", baseUrl: "https://nano-gpt.com/api/v1", adapter: "openai-chat", authKind: "key", dashboardUrl: "https://nano-gpt.com/api" },
@@ -430,7 +417,6 @@ export const PROVIDER_REGISTRY: readonly ProviderRegistryEntry[] = [
     adapter: "openai-chat",
     authKind: "key",
     dashboardUrl: "https://ollama.com/settings/keys",
-    reasoningEffortMap: OLLAMA_REASONING_MAP,
     models: ["glm-5.2", "deepseek-v4-pro", "qwen3-coder", "gpt-oss:120b", "kimi-k2.6", "minimax-m3", "qwen3.5", "gemma4"],
     defaultModel: "glm-5.2",
     noVisionModels: [

package/src/reasoning-effort.ts

@@ -6,6 +6,7 @@ export const CODEX_REASONING_LEVELS: { effort: string; description: string }[] =
   { effort: "medium", description: "Balances speed and reasoning depth" },
   { effort: "high", description: "Greater reasoning depth for complex problems" },
   { effort: "xhigh", description: "Extended reasoning for the hardest problems" },
+  { effort: "max", description: "Maximum reasoning for the hardest problems" },
 ];
 
 const CODEX_REASONING_ORDER = CODEX_REASONING_LEVELS.map(l => l.effort);
@@ -53,7 +54,6 @@ export function configuredReasoningEfforts(provider: OcxProviderConfig, modelId:
 function requestToCodexEffort(requested: string): string | undefined {
   if (requested === "none") return undefined;
   if (requested === "minimal") return "low";
-  if (requested === "max") return "xhigh";
   return CODEX_REASONING_SET.has(requested) ? requested : undefined;
 }
 
@@ -82,9 +82,8 @@ export function reasoningEffortMapFor(provider: OcxProviderConfig, modelId: stri
 }
 
 /**
- * Translate Codex's reasoning label into the provider's real wire value. The Codex catalog must only
- * advertise labels Codex itself accepts (`low`/`medium`/`high`/`xhigh`), but some upstreams use
- * different values (`max`) or a smaller subset (`low`/`medium`/`high`).
+ * Translate Codex's reasoning label into the provider's real wire value. Prefer identity labels
+ * (`xhigh` stays `xhigh`, `max` stays `max`); provider maps are only for real upstream aliases.
  */
 export function mapReasoningEffort(provider: OcxProviderConfig, modelId: string, requested: string | undefined): string | undefined {
   if (!requested) return undefined;

package/src/router.ts

@@ -29,9 +29,8 @@ const MODEL_PROVIDER_PATTERNS: Array<{ providerNames: string[]; prefixes: string
   },
 ];
 
-// Merge registry-default effort maps under user values so persisted built-in provider configs
-// that predate reasoningEffortMap/modelReasoningEffortMap still get correct wire translations
-// (e.g. ollama-cloud xhigh -> max) without a disk migration. User overrides win per-key.
+// Merge registry-default effort maps under user values so built-in provider configs can
+// carry real upstream aliases without a disk migration. User overrides win per-key.
 function mergeRecord(
   seed: Record<string, string> | undefined,
   user: Record<string, string> | undefined,

package/src/server/request-log.ts

@@ -159,12 +159,12 @@ export function usageFromResponsesPayload(usage: unknown): OcxUsage | undefined
   const raw = usage as {
     input_tokens?: unknown;
     output_tokens?: unknown;
-    input_tokens_details?: { cached_tokens?: unknown };
+    input_tokens_details?: { cached_tokens?: unknown; cache_write_tokens?: unknown };
     output_tokens_details?: { reasoning_tokens?: unknown };
     total_tokens?: unknown;
     prompt_tokens?: unknown;
     completion_tokens?: unknown;
-    prompt_tokens_details?: { cached_tokens?: unknown };
+    prompt_tokens_details?: { cached_tokens?: unknown; cache_write_tokens?: unknown };
     completion_tokens_details?: { reasoning_tokens?: unknown };
   };
   if (typeof raw.input_tokens === "number" && typeof raw.output_tokens === "number") {
@@ -175,6 +175,9 @@ export function usageFromResponsesPayload(usage: unknown): OcxUsage | undefined
       ...(typeof raw.input_tokens_details?.cached_tokens === "number"
         ? { cachedInputTokens: raw.input_tokens_details.cached_tokens }
         : {}),
+      ...(typeof raw.input_tokens_details?.cache_write_tokens === "number"
+        ? { cacheCreationInputTokens: raw.input_tokens_details.cache_write_tokens }
+        : {}),
       ...(typeof raw.output_tokens_details?.reasoning_tokens === "number"
         ? { reasoningOutputTokens: raw.output_tokens_details.reasoning_tokens }
         : {}),
@@ -188,6 +191,9 @@ export function usageFromResponsesPayload(usage: unknown): OcxUsage | undefined
       ...(typeof raw.prompt_tokens_details?.cached_tokens === "number"
         ? { cachedInputTokens: raw.prompt_tokens_details.cached_tokens }
         : {}),
+      ...(typeof raw.prompt_tokens_details?.cache_write_tokens === "number"
+        ? { cacheCreationInputTokens: raw.prompt_tokens_details.cache_write_tokens }
+        : {}),
       ...(typeof raw.completion_tokens_details?.reasoning_tokens === "number"
         ? { reasoningOutputTokens: raw.completion_tokens_details.reasoning_tokens }
         : {}),

package/src/types.ts

@@ -322,6 +322,12 @@ export interface OcxTokenGuardianConfig {
   failureBackoffBaseSeconds?: number;
   /** Backoff ceiling (seconds). Default 3600. */
   failureBackoffMaxSeconds?: number;
+  /** Optional Codex pool session warmup sweep. Default false to avoid background synthetic traffic. */
+  codexWarmupEnabled?: boolean;
+  /** Max age before a Codex pool account is revalidated via `/codex/responses`. Default 691200 (8d). */
+  codexWarmupMaxAgeSeconds?: number;
+  /** Model used for optional Codex pool warmup. Default gpt-5.4-mini. */
+  codexWarmupModel?: string;
 }
 
 export interface OcxVisionSidecarConfig {
@@ -396,7 +402,7 @@ export interface OcxProviderConfig {
   refreshPolicy?: RefreshPolicy;
   /**
    * Provider-wide Codex-visible reasoning tiers for routed models. Use only Codex-supported labels
-   * here (`low`, `medium`, `high`, `xhigh`); translate to provider-specific wire values with
+   * here (`low`, `medium`, `high`, `xhigh`, `max`); translate provider aliases with
    * `reasoningEffortMap` / `modelReasoningEffortMap` below.
    */
   reasoningEfforts?: string[];
@@ -489,4 +495,7 @@ export interface CodexAccountCredentialRecord {
   refreshGrantFingerprint?: string;
   deletedAt?: number;
   replacedAt?: number;
+  lastCodexValidatedAt?: number;
+  lastCodexValidationStatus?: "ok" | "failed";
+  lastCodexValidationError?: string;
 }