diff --git a/docs/models.md b/docs/models.md
index 0c4c95d..d4dfa0b 100644
--- a/docs/models.md
+++ b/docs/models.md
@@ -34,6 +34,8 @@ The catalog also lists model ids by provider or endpoint. `/login` and `/model`
 
 Saved provider profiles carry an `extension` tag. Existing profiles without one migrate as `openai`. Codex profiles use `extension: "codex"` and keep OAuth credentials separately in `$ALBEDO_HOME/auth.json`; see [model authentication](auth.md).
 
+Codex metadata lookups select the `openai` provider in models.dev explicitly, including its reasoning effort tiers and context limits. `/effort`, session defaults, and model switches use this same selection; another provider publishing the same model id cannot replace OpenAI's metadata. No authenticated Codex model catalog is fetched.
+
 ## antigravity models
 
 Antigravity is not on models.dev, and its model ids and required `model_enum` labels change server-side. The `antigravity` extension therefore contributes its own catalog. It asks Cloud Code Assist `fetchAvailableModels` which models the account can use, keeps the ids that the real client's agent picker shows, and caches them with their limits, labels, and the renames of retired ids in `$ALBEDO_HOME/antigravity.json`. The cache refreshes in the background after 6 hours, and the first `/login` listing waits for discovery. A built-in table supplies each model's thinking control and serves as the list before the first discovery. A saved profile that names a retired id follows the backend's rename. The catalog answers lookups only for the Antigravity endpoint, so models.dev keeps answering for the same ids at other providers.
diff --git a/src/albedo/daemon/server.gleam b/src/albedo/daemon/server.gleam
index c5d3e71..726df0a 100644
--- a/src/albedo/daemon/server.gleam
+++ b/src/albedo/daemon/server.gleam
@@ -9,6 +9,7 @@ import albedo/daemon/images
 import albedo/daemon/mail
 import albedo/daemon/reaper
 import albedo/daemon/session
+import albedo/daemon/session_provider
 import albedo/daemon/store
 import albedo/daemon/usage
 import albedo/harness/command
@@ -157,14 +158,13 @@ fn handle(state: State, message: Message) {
           "" -> provider.model
           _ -> model
         }
-        let efforts = case runtime.global(state.host) {
-          Ok(extensions) ->
-            case extension.model_info(extensions, model, "") {
-              Some(m_info) -> m_info.efforts
-              None -> []
-            }
-          Error(_) -> []
-        }
+        let efforts =
+          session_provider.model_efforts(
+            state.host,
+            state.config.home,
+            provider.name,
+            model,
+          )
         let effort = extension.default_effort(efforts)
         let info =
           conversation.Info(
diff --git a/src/albedo/daemon/session.gleam b/src/albedo/daemon/session.gleam
index 47ac18e..c0e42f5 100644
--- a/src/albedo/daemon/session.gleam
+++ b/src/albedo/daemon/session.gleam
@@ -173,14 +173,8 @@ pub fn start(
     let info = case info.effort {
       Some(_) -> info
       None -> {
-        let efforts = case runtime.global(host) {
-          Ok(extensions) ->
-            case extension.model_info(extensions, info.model, "") {
-              Some(m_info) -> m_info.efforts
-              None -> []
-            }
-          Error(_) -> []
-        }
+        let efforts =
+          session_provider.model_efforts(host, home, info.provider, info.model)
         case extension.default_effort(efforts) {
           Some(def) -> {
             let _ =
diff --git a/src/albedo/daemon/session_provider.gleam b/src/albedo/daemon/session_provider.gleam
index fb981a1..9c7f24e 100644
--- a/src/albedo/daemon/session_provider.gleam
+++ b/src/albedo/daemon/session_provider.gleam
@@ -57,9 +57,23 @@ pub fn efforts_for_model(
   state: session_state.State(message),
   model: String,
 ) -> List(String) {
-  case runtime.global(state.host) {
+  model_efforts(state.host, state.home, state.info.provider, model)
+}
+
+pub fn model_efforts(
+  host: runtime.Runtime,
+  home: String,
+  provider: String,
+  model: String,
+) -> List(String) {
+  let endpoint = case configuration.named(home, provider) {
+    Ok(configured) if configured.extension == "codex" ->
+      "https://chatgpt.com/backend-api"
+    _ -> ""
+  }
+  case runtime.global(host) {
     Ok(extensions) ->
-      case extension.model_info(extensions, model, "") {
+      case extension.model_info(extensions, model, endpoint) {
         option.Some(info) -> info.efforts
         option.None -> []
       }
@@ -188,7 +202,7 @@ pub fn select(
                   #(configured.name, configured.protocol)
                 })
             })
-            let efforts = efforts_for_model(state, model)
+            let efforts = model_efforts(state.host, state.home, provider, model)
             let new_effort = case state.info.effort {
               option.Some(current) ->
                 case list.contains(efforts, current) {
diff --git a/src/albedo/harness/extensions/models/albedo_models.erl b/src/albedo/harness/extensions/models/albedo_models.erl
index 217d53e..328cc44 100644
--- a/src/albedo/harness/extensions/models/albedo_models.erl
+++ b/src/albedo/harness/extensions/models/albedo_models.erl
@@ -325,6 +325,13 @@ resolve(#{index := Index, providers := Providers}, Model, Host) ->
 
 %% The configured endpoint decides between providers that publish one model id.
 %% Without a host match, agreeing candidates still answer and conflicting ones do not.
+%% Codex subscription models use OpenAI's metadata, even though models.dev
+%% does not publish the ChatGPT endpoint (or an API URL for OpenAI).
+select(Candidates, _Providers, <<"chatgpt.com">>) ->
+    case [E || {<<"openai">>, _} = E <- Candidates] of
+        [Entry | _] -> {ok, Entry, <<"provider identity">>};
+        [] -> error
+    end;
 select(Candidates, Providers, Host) ->
     case [E || {Name, _} = E <- Candidates, Host =/= <<>>,
                element(1, maps:get(Name, Providers)) =:= Host] of