diff --git a/CUSTOMIZATION.md b/CUSTOMIZATION.md
index 74d507de..2573feab 100644
--- a/CUSTOMIZATION.md
+++ b/CUSTOMIZATION.md
@@ -5,7 +5,7 @@ Open SWE is designed to be forked and customized for your org. The core agent is
 ```python
 # agent/server.py — the key lines
 return create_deep_agent(
-    model=make_model("anthropic:claude-opus-4-6", temperature=0, max_tokens=20_000),
+    model=make_model(os.environ.get("LLM_MODEL_ID", DEFAULT_LLM_MODEL_ID), temperature=0, max_tokens=20_000),
     system_prompt=construct_system_prompt(repo_dir, ...),
     tools=[http_request, fetch_url, commit_and_open_pr, linear_comment, slack_thread_reply],
     backend=sandbox_backend,
@@ -117,12 +117,15 @@ See `agent/integrations/langsmith.py` (`LangSmithBackend` class) for a full refe
 
 ## 2. Model
 
-The model is configured in the `get_agent()` function in `agent/server.py`:
+The model is configured in the `get_agent()` function in `agent/server.py`. By default it uses `anthropic:claude-opus-4-6`, but you can override it with the `LLM_MODEL_ID` environment variable:
 
-```python
-model=make_model("anthropic:claude-opus-4-6", temperature=0, max_tokens=20_000)
+```bash
+# Set the model via environment variable (uses provider:model format)
+LLM_MODEL_ID="anthropic:claude-sonnet-4-6"
 ```
 
+If `LLM_MODEL_ID` is not set, the default model (`anthropic:claude-opus-4-6`) is used.
+
 ### Switching models
 
 Use the `provider:model` format:
diff --git a/agent/server.py b/agent/server.py
index e4fa208b..10b4e714 100644
--- a/agent/server.py
+++ b/agent/server.py
@@ -4,6 +4,7 @@
 # Suppress deprecation warnings from langchain_core (e.g., Pydantic V1 on Python 3.14+)
 # ruff: noqa: E402
 import logging
+import os
 import shlex
 import warnings
 
@@ -244,6 +245,7 @@ def graph_loaded_for_execution(config: RunnableConfig) -> bool:
     )
 
 
+DEFAULT_LLM_MODEL_ID = "anthropic:claude-opus-4-6"
 DEFAULT_RECURSION_LIMIT = 1_000
 
 
@@ -400,7 +402,11 @@ async def get_agent(config: RunnableConfig) -> Pregel:  # noqa: PLR0915
 
     logger.info("Returning agent with sandbox for thread %s", thread_id)
     return create_deep_agent(
-        model=make_model("anthropic:claude-opus-4-6", temperature=0, max_tokens=20_000),
+        model=make_model(
+            os.environ.get("LLM_MODEL_ID", DEFAULT_LLM_MODEL_ID),
+            temperature=0,
+            max_tokens=20_000,
+        ),
         system_prompt=construct_system_prompt(
             repo_dir,
             linear_project_id=linear_project_id,