diff --git a/codex-rs/core/src/tools/handlers/collab.rs b/codex-rs/core/src/tools/handlers/collab.rs
index 56463e644a..338f18df91 100644
--- a/codex-rs/core/src/tools/handlers/collab.rs
+++ b/codex-rs/core/src/tools/handlers/collab.rs
@@ -13,6 +13,7 @@ use crate::tools::registry::ToolKind;
 use async_trait::async_trait;
 use codex_protocol::ThreadId;
 use codex_protocol::models::BaseInstructions;
+use codex_protocol::openai_models::ReasoningEffort;
 use codex_protocol::protocol::CollabAgentInteractionBeginEvent;
 use codex_protocol::protocol::CollabAgentInteractionEndEvent;
 use codex_protocol::protocol::CollabAgentSpawnBeginEvent;
@@ -84,6 +85,8 @@ mod spawn {
     struct SpawnAgentArgs {
         message: String,
         agent_type: Option<AgentRole>,
+        model: Option<String>,
+        reasoning_effort: Option<ReasoningEffort>,
     }
 
     #[derive(Debug, Serialize)]
@@ -99,6 +102,13 @@ mod spawn {
     ) -> Result<ToolOutput, FunctionCallError> {
         let args: SpawnAgentArgs = parse_arguments(&arguments)?;
         let agent_role = args.agent_type.unwrap_or(AgentRole::Default);
+        if let Some(model) = args.model.as_deref()
+            && model.trim().is_empty()
+        {
+            return Err(FunctionCallError::RespondToModel(
+                "model must be non-empty when provided".to_string(),
+            ));
+        }
         let prompt = args.message;
         if prompt.trim().is_empty() {
             return Err(FunctionCallError::RespondToModel(
@@ -121,7 +131,12 @@ mod spawn {
         agent_role
             .apply_to_config(&mut config)
             .map_err(FunctionCallError::RespondToModel)?;
+        // Allow explicit per-spawn overrides to supersede both the inherited turn
+        // configuration and any role-specific defaults (e.g. Worker model).
+        apply_agent_spawn_overrides(&mut config, args.model, args.reasoning_effort);
 
+        let model = config.model.clone();
+        let reasoning_effort = config.model_reasoning_effort;
         let result = session
             .services
             .agent_control
@@ -143,6 +158,8 @@ mod spawn {
                     sender_thread_id: session.conversation_id,
                     new_thread_id,
                     prompt,
+                    model,
+                    reasoning_effort,
                     status,
                 }
                 .into(),
@@ -592,6 +609,19 @@ fn build_agent_spawn_config(
     Ok(config)
 }
 
+fn apply_agent_spawn_overrides(
+    config: &mut Config,
+    model: Option<String>,
+    reasoning_effort: Option<ReasoningEffort>,
+) {
+    if let Some(model) = model {
+        config.model = Some(model);
+    }
+    if let Some(reasoning_effort) = reasoning_effort {
+        config.model_reasoning_effort = Some(reasoning_effort);
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -723,6 +753,54 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn spawn_agent_rejects_empty_model_override() {
+        let (session, turn) = make_session_and_context().await;
+        let invocation = invocation(
+            Arc::new(session),
+            Arc::new(turn),
+            "spawn_agent",
+            function_payload(json!({
+                "message": "hello",
+                "model": "   ",
+            })),
+        );
+        let Err(err) = CollabHandler.handle(invocation).await else {
+            panic!("empty model should be rejected");
+        };
+        assert_eq!(
+            err,
+            FunctionCallError::RespondToModel("model must be non-empty when provided".to_string())
+        );
+    }
+
+    #[tokio::test]
+    async fn spawn_overrides_win_over_role_defaults() {
+        use crate::agent::AgentRole;
+
+        let (_session, turn) = make_session_and_context().await;
+        let base_instructions = BaseInstructions {
+            text: "base".to_string(),
+        };
+
+        let mut config = build_agent_spawn_config(&base_instructions, &turn).expect("spawn config");
+        AgentRole::Worker
+            .apply_to_config(&mut config)
+            .expect("apply worker role");
+
+        apply_agent_spawn_overrides(
+            &mut config,
+            Some("gpt-5".to_string()),
+            Some(ReasoningEffort::Minimal),
+        );
+
+        assert_eq!(config.model, Some("gpt-5".to_string()));
+        assert_eq!(
+            config.model_reasoning_effort,
+            Some(ReasoningEffort::Minimal)
+        );
+    }
+
     #[tokio::test]
     async fn send_input_rejects_empty_message() {
         let (session, turn) = make_session_and_context().await;
diff --git a/codex-rs/core/src/tools/spec.rs b/codex-rs/core/src/tools/spec.rs
index e330d93a03..631acbf28a 100644
--- a/codex-rs/core/src/tools/spec.rs
+++ b/codex-rs/core/src/tools/spec.rs
@@ -454,6 +454,24 @@ fn create_spawn_agent_tool() -> ToolSpec {
             )),
         },
     );
+    properties.insert(
+        "model".to_string(),
+        JsonSchema::String {
+            description: Some(
+                "Optional model override for the spawned agent (e.g. \"gpt-5.2-codex\"). When set, overrides the inherited session model and any agent_type defaults."
+                    .to_string(),
+            ),
+        },
+    );
+    properties.insert(
+        "reasoning_effort".to_string(),
+        JsonSchema::String {
+            description: Some(
+                "Optional reasoning effort for the spawned agent (one of: \"none\", \"minimal\", \"low\", \"medium\", \"high\", \"xhigh\"). When set, overrides the inherited session reasoning effort."
+                    .to_string(),
+            ),
+        },
+    );
 
     ToolSpec::Function(ResponsesApiTool {
         name: "spawn_agent".to_string(),
diff --git a/codex-rs/core/templates/agents/orchestrator.md b/codex-rs/core/templates/agents/orchestrator.md
index 09b6ea0b4e..b66df72418 100644
--- a/codex-rs/core/templates/agents/orchestrator.md
+++ b/codex-rs/core/templates/agents/orchestrator.md
@@ -53,9 +53,37 @@ You are Codex Orchestrator, based on GPT-5. You are running as an orchestration
 * By default, workers must not spawn sub-agents unless explicitly allowed.
 * When multiple workers are active, you may pass multiple IDs to `wait` to react to the first completion and keep the workflow event-driven and use a long timeout (e.g. 5 minutes).
 
+## Choosing model and reasoning effort
+
+When spawning a worker, you may optionally override its `model` and/or `reasoning_effort`. If provided, these overrides take precedence over the inherited session settings and any `agent_type` defaults.
+
+### Models (OpenAI)
+
+Use the most capable model that matches the task and latency/cost constraints:
+
+* `gpt-5.2-codex` (default): best general choice for repo-scale coding, tool use, and multi-step changes.
+* `gpt-5.1-codex-max`: strong for deep reasoning on tricky code, large diffs, and higher-risk refactors.
+* `gpt-5.1-codex-mini`: faster/cheaper; good for quick triage, small edits, and straightforward tasks.
+* `gpt-5.2`: strong general model when the work is less tool-heavy and needs broader knowledge/explanations.
+* `gpt-5`: general model with broad knowledge; use for documentation, reasoning, or non-coding-heavy sub-tasks.
+* `gpt-5.1`: older general model; use only when you need parity with prior behavior.
+
+### Reasoning effort
+
+Pick the smallest effort that is likely to succeed on the first try:
+
+* `minimal`: trivial lookups, short explanations, mechanical edits.
+* `low`: straightforward tasks with clear requirements; fast iteration.
+* `medium`: default for typical coding/debugging work.
+* `high`: ambiguous tasks, multi-file changes, tricky debugging, integration work.
+* `xhigh`: large/complex refactors, deep investigations, or when prior attempts failed.
+* `none`: rarely; only when you explicitly want to suppress extra reasoning overhead.
+
+Note: not every model supports every effort. If an effort isn't supported, choose the closest supported level (e.g. `high` instead of `xhigh`).
+
 ## Collab tools
 
-* `spawn_agent`: create a worker with an initial prompt (`agent_type` required).
+* `spawn_agent`: create a worker with an initial prompt (`agent_type` optional; `model` and `reasoning_effort` optional overrides).
 * `send_input`: send follow-ups or fixes (queued unless interrupted).
 * `send_input(interrupt=true)`: stop current work and redirect immediately.
 * `wait`: wait for one or more workers; returns when at least one finishes.
diff --git a/codex-rs/protocol/src/protocol.rs b/codex-rs/protocol/src/protocol.rs
index 1f35fed60e..13b316bfde 100644
--- a/codex-rs/protocol/src/protocol.rs
+++ b/codex-rs/protocol/src/protocol.rs
@@ -2243,6 +2243,12 @@ pub struct CollabAgentSpawnEndEvent {
     /// Initial prompt sent to the agent. Can be empty to prevent CoT leaking at the
     /// beginning.
     pub prompt: String,
+    /// Model slug used for the newly spawned agent.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub model: Option<String>,
+    /// Reasoning effort used for the newly spawned agent.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_effort: Option<ReasoningEffortConfig>,
     /// Last known status of the new agent reported to the sender agent.
     pub status: AgentStatus,
 }
diff --git a/codex-rs/tui/src/collab.rs b/codex-rs/tui/src/collab.rs
index b6c7d80959..5b3909897b 100644
--- a/codex-rs/tui/src/collab.rs
+++ b/codex-rs/tui/src/collab.rs
@@ -23,6 +23,8 @@ pub(crate) fn spawn_end(ev: CollabAgentSpawnEndEvent) -> PlainHistoryCell {
         sender_thread_id: _,
         new_thread_id,
         prompt,
+        model,
+        reasoning_effort,
         status,
     } = ev;
     let new_agent = new_thread_id
@@ -31,8 +33,14 @@ pub(crate) fn spawn_end(ev: CollabAgentSpawnEndEvent) -> PlainHistoryCell {
     let mut details = vec![
         detail_line("call", call_id),
         detail_line("agent", new_agent),
-        status_line(&status),
     ];
+    if let Some(model) = model {
+        details.push(detail_line("model", model));
+    }
+    if let Some(reasoning_effort) = reasoning_effort {
+        details.push(detail_line("effort", reasoning_effort.to_string()));
+    }
+    details.push(status_line(&status));
     if let Some(line) = prompt_line(&prompt) {
         details.push(line);
     }