deepagents: update default model (#3575)

Author: ccurme

src/code-samples/deepagents/content-builder.py

@@ -140,7 +140,7 @@ def load_subagents(config_path: Path) -> list:
 def create_content_writer():
     """Create a content writer agent configured by filesystem files."""
     return create_deep_agent(
-        model="openai:gpt-5.4",
+        model="anthropic:claude-sonnet-4-6",
         memory=["./AGENTS.md"],
         skills=["./skills/"],
         tools=[generate_cover, generate_social_image],

src/langsmith/trace-deep-agents.mdx

@@ -103,7 +103,7 @@ def yearly_balance_schedule(
 
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     tools=[compute_compound_interest, yearly_balance_schedule],
     system_prompt=(
         "You are a careful assistant. "
@@ -239,7 +239,7 @@ def yearly_balance_schedule(
 
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     tools=[compute_compound_interest, yearly_balance_schedule],
     system_prompt=(
         "You are a careful assistant. "

src/oss/deepagents/acp.mdx

@@ -59,7 +59,7 @@ from deepagents_acp.server import AgentServerACP
 
 async def main() -> None:
     agent = create_deep_agent(
-        model="openai:gpt-5.4",
+        model="google_genai:gemini-3.1-pro-preview",
         # You can customize your Deep Agent here: set a custom prompt,
         # add your own tools, attach middleware, or compose subagents.
         system_prompt="You are a helpful coding assistant",

src/oss/deepagents/async-subagents.mdx

@@ -67,7 +67,7 @@ async_subagents = [
 ]
 
 agent = create_deep_agent(
-    model="claude-sonnet-4-6",
+    model="google_genai:gemini-3.1-pro-preview",
     subagents=async_subagents,
 )
 ```
@@ -101,7 +101,7 @@ const asyncSubagents: AsyncSubAgent[] = [
 ];
 
 const agent = createDeepAgent({
-  model: "claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   subagents: [...asyncSubagents],
 });
 ```
@@ -348,7 +348,7 @@ When using LangGraph-based deployments, every async subagent run is a standard L
 :::python
 ```python
 agent = create_deep_agent(
-    model="claude-sonnet-4-6",
+    model="google_genai:gemini-3.1-pro-preview",
     system_prompt="""...your instructions...
 
     After launching an async subagent, ALWAYS return control to the user.
@@ -361,7 +361,7 @@ agent = create_deep_agent(
 :::js
 ```typescript
 const agent = createDeepAgent({
-  model: "claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   systemPrompt: `...your instructions...
 
     After launching an async subagent, ALWAYS return control to the user.

src/oss/deepagents/backends.mdx

@@ -43,11 +43,11 @@ Here are a few prebuilt filesystem backends that you can quickly use with your d
 
 | Built-in backend | Description |
 |---|---|
-| [Default](#statebackend-ephemeral) | `agent = create_deep_agent(model="openai:gpt-5.4")` <br></br> Ephemeral in state. The default filesystem backend for an agent is stored in `langgraph` state. Note that this filesystem only persists _for a single thread_. |
-| [Local filesystem persistence](#filesystembackend-local-disk) | `agent = create_deep_agent(model="openai:gpt-5.4", backend=FilesystemBackend(root_dir="/Users/nh/Desktop/"))` <br></br>This gives the deep agent access to your local machine's filesystem. You can specify the root directory that the agent has access to. Note that any provided `root_dir` must be an absolute path. |
-| [Durable store (LangGraph store)](#storebackend-langgraph-store) | `agent = create_deep_agent(model="openai:gpt-5.4", backend=StoreBackend())` <br></br>This gives the agent access to long-term storage that is _persisted across threads_. This is great for storing longer term memories or instructions that are applicable to the agent over multiple executions. |
-| [Sandbox](/oss/deepagents/sandboxes) | `agent = create_deep_agent(model="openai:gpt-5.4", backend=sandbox)` <br></br>Execute code in isolated environments. Sandboxes provide filesystem tools plus the `execute` tool for running shell commands. Choose from Modal, Daytona, Deno, or local VFS. |
-| [Local shell](#localshellbackend-local-shell) | `agent = create_deep_agent(model="openai:gpt-5.4", backend=LocalShellBackend(root_dir=".", env={"PATH": "/usr/bin:/bin"}))` <br></br>Filesystem and shell execution directly on the host. No isolation—use only in controlled development environments. See [security considerations](#localshellbackend-local-shell) below. |
+| [Default](#statebackend-ephemeral) | `agent = create_deep_agent(model="google_genai:gemini-3.1-pro-preview")` <br></br> Ephemeral in state. The default filesystem backend for an agent is stored in `langgraph` state. Note that this filesystem only persists _for a single thread_. |
+| [Local filesystem persistence](#filesystembackend-local-disk) | `agent = create_deep_agent(model="google_genai:gemini-3.1-pro-preview", backend=FilesystemBackend(root_dir="/Users/nh/Desktop/"))` <br></br>This gives the deep agent access to your local machine's filesystem. You can specify the root directory that the agent has access to. Note that any provided `root_dir` must be an absolute path. |
+| [Durable store (LangGraph store)](#storebackend-langgraph-store) | `agent = create_deep_agent(model="google_genai:gemini-3.1-pro-preview", backend=StoreBackend())` <br></br>This gives the agent access to long-term storage that is _persisted across threads_. This is great for storing longer term memories or instructions that are applicable to the agent over multiple executions. |
+| [Sandbox](/oss/deepagents/sandboxes) | `agent = create_deep_agent(model="google_genai:gemini-3.1-pro-preview", backend=sandbox)` <br></br>Execute code in isolated environments. Sandboxes provide filesystem tools plus the `execute` tool for running shell commands. Choose from Modal, Daytona, Deno, or local VFS. |
+| [Local shell](#localshellbackend-local-shell) | `agent = create_deep_agent(model="google_genai:gemini-3.1-pro-preview", backend=LocalShellBackend(root_dir=".", env={"PATH": "/usr/bin:/bin"}))` <br></br>Filesystem and shell execution directly on the host. No isolation—use only in controlled development environments. See [security considerations](#localshellbackend-local-shell) below. |
 | [Composite](#compositebackend-router) | Ephemeral by default, `/memories/` persisted. The Composite backend is maximally flexible. You can specify different routes in the filesystem to point towards different backends. See Composite routing below for a ready-to-paste example. |
 
 
@@ -369,7 +369,7 @@ from deepagents import create_deep_agent
 from deepagents.backends import CompositeBackend, StateBackend, FilesystemBackend
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     backend=CompositeBackend(
         default=StateBackend(),
         routes={
@@ -555,7 +555,7 @@ Use [permissions](/oss/deepagents/permissions) to declaratively control which fi
 from deepagents import create_deep_agent, FilesystemPermission
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     backend=CompositeBackend(
         default=StateBackend(),
         routes={
@@ -828,7 +828,7 @@ from deepagents import create_deep_agent
 from deepagents.backends import CompositeBackend, StateBackend, StoreBackend
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     backend=lambda rt: CompositeBackend(
         default=StateBackend(rt),
         routes={"/memories/": StoreBackend(rt, namespace=lambda rt: (rt.server_info.user.identity,))},
@@ -837,7 +837,7 @@ agent = create_deep_agent(
 
 # After
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     backend=CompositeBackend(
         default=StateBackend(),
         routes={"/memories/": StoreBackend(namespace=lambda rt: (rt.server_info.user.identity,))},

src/oss/deepagents/context-engineering.mdx

@@ -54,7 +54,7 @@ Your custom system prompt is prepended to the built-in system prompt, which incl
 from deepagents import create_deep_agent
 
 agent = create_deep_agent(
-    model="claude-sonnet-4-6",
+    model="google_genai:gemini-3.1-pro-preview",
     system_prompt=(
         "You are a research assistant specializing in scientific literature. "
         "Always cite sources. Use subagents for parallel research on different topics."
@@ -68,7 +68,7 @@ agent = create_deep_agent(
 import { createDeepAgent } from "deepagents";
 
 const agent = await createDeepAgent({
-  model: "claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   systemPrompt: `You are a research assistant specializing in scientific literature.
   Always cite sources. Use subagents for parallel research on different topics.`,
 });
@@ -101,7 +101,7 @@ Memory files ([`AGENTS.md`](https://agents.md/)) provide persistent context that
 :::python
 ```python
 agent = create_deep_agent(
-    model="claude-sonnet-4-6",
+    model="google_genai:gemini-3.1-pro-preview",
     memory=["/project/AGENTS.md", "~/.deepagents/preferences.md"],
 )
 ```
@@ -110,7 +110,7 @@ agent = create_deep_agent(
 :::js
 ```typescript
 const agent = await createDeepAgent({
-  model: "claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   memory: ["/project/AGENTS.md", "~/.deepagents/preferences.md"],
 });
 ```
@@ -125,7 +125,7 @@ Skills provide **on-demand** capabilities. The agent reads frontmatter from each
 :::python
 ```python
 agent = create_deep_agent(
-    model="claude-sonnet-4-6",
+    model="google_genai:gemini-3.1-pro-preview",
     skills=["/skills/research/", "/skills/web-search/"],
 )
 ```
@@ -134,7 +134,7 @@ agent = create_deep_agent(
 :::js
 ```typescript
 const agent = await createDeepAgent({
-  model: "claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   skills: ["/skills/research/", "/skills/web-search/"],
 });
 ```
@@ -243,7 +243,7 @@ def fetch_user_data(query: str, runtime: ToolRuntime[Context]) -> str:
     return f"Data for user {user_id}: {query}"
 
 agent = create_deep_agent(
-    model="claude-sonnet-4-6",
+    model="google_genai:gemini-3.1-pro-preview",
     tools=[fetch_user_data],
     context_schema=Context,
 )
@@ -284,7 +284,7 @@ const fetchUserData = tool(
 );
 
 const agent = await createDeepAgent({
-  model: "claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   tools: [fetchUserData],
   contextSchema,
 });
@@ -365,7 +365,7 @@ from deepagents.middleware.summarization import (
 
 backend = StateBackend  # if using default backend
 
-model = "openai:gpt-5.4"
+model = "google_genai:gemini-3.1-pro-preview"
 agent = create_deep_agent(
     model=model,
     middleware=[  # [!code highlight]
@@ -449,7 +449,7 @@ def make_backend(runtime):
     )
 
 agent = create_deep_agent(
-    model="claude-sonnet-4-6",
+    model="google_genai:gemini-3.1-pro-preview",
     store=InMemoryStore(),
     backend=make_backend,
     system_prompt="""When users tell you their preferences, save them to
@@ -464,7 +464,7 @@ import { createDeepAgent, CompositeBackend, StateBackend, StoreBackend } from "d
 import { InMemoryStore } from "@langchain/langgraph-checkpoint";
 
 const agent = await createDeepAgent({
-  model: "claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   store: new InMemoryStore(),
   backend: new CompositeBackend(
     new StateBackend(),

src/oss/deepagents/customization.mdx

@@ -119,7 +119,7 @@ from deepagents import create_deep_agent
 
 agent = create_deep_agent(
     model=init_chat_model(
-        model="claude-sonnet-4-6",
+        model="google_genai:gemini-3.1-pro-preview",
         max_retries=10,  # Increase for unreliable networks (default: 6)
         timeout=120,     # Increase timeout for slow connections
     ),
@@ -177,7 +177,7 @@ def internet_search(
     )
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     tools=[internet_search]
 )
 ```
@@ -248,7 +248,7 @@ thorough research, and then write a polished report. \
 """
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     system_prompt=research_instructions,
 )
 ```
@@ -343,7 +343,7 @@ def log_tool_calls(request, handler):
 
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     tools=[get_weather],
     middleware=[log_tool_calls],
 )
@@ -394,7 +394,7 @@ const logToolCallsMiddleware = createMiddleware({
 });
 
 const agent = await createDeepAgent({
-  model: "anthropic:claude-sonnet-4-6",
+  model: "google_genai:gemini-3.1-pro-preview",
   tools: [getWeather] as any,
   middleware: [logToolCallsMiddleware] as any,
 });
@@ -644,7 +644,7 @@ You can pass one or more file paths to the `memory` parameter when creating your
         checkpointer = MemorySaver()
 
         agent = create_deep_agent(
-            model="openai:gpt-5.4",
+            model="google_genai:gemini-3.1-pro-preview",
             memory=[
                 "/AGENTS.md"
             ],
@@ -688,7 +688,7 @@ You can pass one or more file paths to the `memory` parameter when creating your
         )
 
         agent = create_deep_agent(
-            model="openai:gpt-5.4",
+            model="google_genai:gemini-3.1-pro-preview",
             backend=StoreBackend(),
             store=store,
             memory=[
@@ -720,7 +720,7 @@ You can pass one or more file paths to the `memory` parameter when creating your
         checkpointer = MemorySaver()
 
         agent = create_deep_agent(
-            model="openai:gpt-5.4",
+            model="google_genai:gemini-3.1-pro-preview",
             backend=FilesystemBackend(root_dir="/Users/user/{project}"),
             memory=[
                 "./AGENTS.md"
@@ -912,7 +912,7 @@ class WeatherReport(BaseModel):
 
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     response_format=WeatherReport,
     tools=[internet_search]
 )
@@ -1013,3 +1013,4 @@ console.log(result.structuredResponse);
 :::
 
 For more information and examples, see [response format](/oss/langchain/structured-output#response-format).
+n and examples, see [response format](/oss/langchain/structured-output#response-format).

src/oss/deepagents/data-analysis.mdx

@@ -264,7 +264,7 @@ from deepagents import create_deep_agent
 checkpointer = InMemorySaver()
 
 agent = create_deep_agent(
-    model="anthropic:claude-sonnet-4-5",
+    model="google_genai:gemini-3.1-pro-preview",
     tools=[slack_send_message],
     backend=backend,
     checkpointer=checkpointer,

src/oss/deepagents/deploy.mdx

@@ -83,7 +83,7 @@ This creates the following files:
 | --- | --- |
 | `deepagents.toml` | Agent config — name, model, optional sandbox |
 | `AGENTS.md` | System prompt loaded at session start |
-| `.env` | API key template (`ANTHROPIC_API_KEY`, `LANGSMITH_API_KEY`, etc.) |
+| `.env` | API key template (`GOOGLE_API_KEY`, `LANGSMITH_API_KEY`, etc.) |
 | `mcp.json` | MCP server configuration (empty by default) |
 | `skills/` | Directory for [Agent Skills](https://agentskills.io/), with an example `review` skill |
 
@@ -140,7 +140,7 @@ Core agent identity. For more on model selection and provider configuration, see
 ```toml deepagents.toml
 [agent]
 name = "research-assistant"
-model = "anthropic:claude-sonnet-4-6"
+model = "google_genai:gemini-3.1-pro-preview"
 ```
 
 <Note>
@@ -151,7 +151,7 @@ Skills, MCP servers, and model dependencies are auto-detected from the project l
 
 - **Skills**: the bundler recursively scans `skills/`, skipping hidden dotfiles, and bundles the rest.
 - **MCP servers**: if `mcp.json` exists, it is included in the deployment and [`langchain-mcp-adapters`](https://pypi.org/project/langchain-mcp-adapters/) is added as a dependency. Only HTTP/SSE transports are supported (stdio is rejected at bundle time).
-- **Model dependencies**: the `provider:` prefix in the `model` field determines the required `langchain-*` package (e.g., `anthropic` -> `langchain-anthropic`).
+- **Model dependencies**: the `provider:` prefix in the `model` field determines the required `langchain-*` package (e.g., `google_genai` -> `langchain-google-genai`).
 - **Sandbox dependencies**: the `[sandbox].provider` value maps to its partner package (e.g., `daytona` -> `langchain-daytona`).
 
 ### `[sandbox]`
@@ -222,15 +222,15 @@ A content writing agent that only needs a model and system prompt, with no code
 ```toml deepagents.toml
 [agent]
 name = "deepagents-deploy-content-writer"
-model = "anthropic:claude-sonnet-4-6"
+model = "google_genai:gemini-3.1-pro-preview"
 ```
 
 A coding agent with a LangSmith sandbox for running code:
 
 ```toml deepagents.toml
 [agent]
 name = "deepagents-deploy-coding-agent"
-model = "anthropic:claude-sonnet-4-6"
+model = "google_genai:gemini-3.1-pro-preview"
 
 [sandbox]
 provider = "langsmith"

src/oss/deepagents/frontend/overview.mdx

@@ -45,7 +45,7 @@ graph LR
 from deepagents import create_deep_agent
 
 agent = create_deep_agent(
-    model="openai:gpt-5.4",
+    model="google_genai:gemini-3.1-pro-preview",
     tools=[get_weather],
     system_prompt="You are a helpful assistant",
     subagents=[