404 - model not found error for Agent as a Tool with run_live bidi

[kalikichandu]

🔴 Required Information

Please ensure all items in this section are completed to allow for efficient
triaging. Requests without complete information may be rejected / deprioritized.
If an item is not applicable to you - please mark it as N/A

Describe the Bug:
When an agent to setup as a tool error 404 invalid model is being thrown even though the model is valid. if the agent is setup as a sub agent then it works as expected

Steps to Reproduce:
Please provide a numbered list of steps to reproduce the behavior:

1. Clone BiDi-Demo app from ADK samples - https://github.com/google/adk-samples/tree/main/python/agents/bidi-demo
2. Create an Orchestrator agent which uses Google search as an Agent Tool
3. Run the main.py with entry point as Orchestrator agent
4. Ask any search question then google_search_agent will be triggered and 404 not found error is thrown even though the model id is valid
5. if you switch google_search_agent as sub agent instead of a tool it works with out any error

Expected Behavior:
404 error should not be thrown if an Agent is setup as a tool

Observed Behavior:
ran list models and this model name is valid one with bidi support

google.genai.errors.ClientError: 404 NOT_FOUND. {'error': {'code': 404, 'message': 'models/gemini-2.5-flash-native-audio-preview-12-2025 is not found for API version v1beta, or is not supported for generateContent. Call ListModels to see the list of available models and their supported methods.', 'status': 'NOT_FOUND'}}

Environment Details:

• ADK Library Version (pip show google-adk): 1.25.1
• Desktop OS:** windows
• Python Version (python -V):3.14.2

Model Information:

• Are you using LiteLLM: No
• Which model is being used: gemini-2.5-flash-native-audio-preview-12-2025

---

🟡 Optional Information

Providing this information greatly speeds up the resolution process.

Regression:
N/A

Logs:
Please attach relevant logs. Wrap them in code blocks (```) or attach a
text file.

2026-02-25 21:39:09,486 - google_adk.google.adk.models.google_llm - INFO - Sending out request, model: gemini-2.5-flash-native-audio-preview-12-2025, backend: GoogleLLMVariant.GEMINI_API, stream: False 
2026-02-25 21:39:09,487 - google_adk.google.adk.models.google_llm - DEBUG - 
LLM Request:
-----------------------------------------------------------
System Instruction:
You are a helpful assistant that can search the web.

You are an agent. Your internal name is "google_search_agent".
-----------------------------------------------------------
Config:
{'http_options': {'headers': {'x-goog-api-client': 'google-adk/1.25.1 gl-python/3.14.2', 'user-agent': 'google-adk/1.25.1 gl-python/3.14.2'}}}
-----------------------------------------------------------
Contents:
{"parts":[{"text":"top 3 hospitality news from singapore"}],"role":"user"}
-----------------------------------------------------------
Functions:

-----------------------------------------------------------

2026-02-25 21:39:09,488 - google_genai.models - INFO - AFC is enabled with max remote calls: 10.     
2026-02-25 21:39:09,531 - httpcore.connection - DEBUG - connect_tcp.started host='generativelanguage.googleapis.com' port=443 local_address=None timeout=None socket_options=None
2026-02-25 21:39:09,822 - httpcore.connection - DEBUG - connect_tcp.complete return_value=<httpcore._backends.anyio.AnyIOStream object at 0x000002BC8C159FD0>
2026-02-25 21:39:09,822 - httpcore.connection - DEBUG - start_tls.started ssl_context=<ssl.SSLContext object at 0x000002BC8BF19BA0> server_hostname='generativelanguage.googleapis.com' timeout=None      
2026-02-25 21:39:09,864 - httpcore.connection - DEBUG - start_tls.complete return_value=<httpcore._backends.anyio.AnyIOStream object at 0x000002BC8C169590>
2026-02-25 21:39:09,866 - httpcore.http11 - DEBUG - send_request_headers.started request=<Request [b'POST']>
2026-02-25 21:39:09,867 - httpcore.http11 - DEBUG - send_request_headers.complete
2026-02-25 21:39:09,868 - httpcore.http11 - DEBUG - send_request_body.started request=<Request [b'POST']>
2026-02-25 21:39:09,869 - httpcore.http11 - DEBUG - send_request_body.complete
2026-02-25 21:39:09,869 - httpcore.http11 - DEBUG - receive_response_headers.started request=<Request [b'POST']>
2026-02-25 21:39:09,952 - httpcore.http11 - DEBUG - receive_response_headers.complete return_value=(b'HTTP/1.1', 404, b'Not Found', [(b'Vary', b'Origin'), (b'Vary', b'X-Origin'), (b'Vary', b'Referer'), (b'Content-Type', b'application/json; charset=UTF-8'), (b'Content-Encoding', b'gzip'), (b'Date', b'Wed, 25 Feb 2026 16:09:12 GMT'), (b'Server', b'scaffolding on HTTPServer2'), (b'X-XSS-Protection', b'0'), (b'X-Frame-Options', b'SAMEORIGIN'), (b'X-Content-Type-Options', b'nosniff'), (b'Server-Timing', b'gfet4t7; dur=66'), (b'Alt-Svc', b'h3=":443"; ma=2592000,h3-29=":443"; ma=2592000'), (b'Transfer-Encoding', b'chunked')])
2026-02-25 21:39:09,956 - httpx - INFO - HTTP Request: POST https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-native-audio-preview-12-2025:generateContent "HTTP/1.1 404 Not Found"
2026-02-25 21:39:09,957 - httpcore.http11 - DEBUG - receive_response_body.started request=<Request [b'POST']>

google.genai.errors.ClientError: 404 NOT_FOUND. {'error': {'code': 404, 'message': 'models/gemini-2.5-flash-native-audio-preview-12-2025 is not found for API version v1beta, or is not supported for generateContent. Call ListModels to see the list of available models and their supported methods.', 'status': 'NOT_FOUND'}}

Screenshots / Video:
N/A

Additional Context:
N/A

Minimal Reproduction Code:
Please provide a code snippet or a link to a Gist/repo that isolates the issue.

# setup ADK BIDI sample - https://github.com/google/adk-samples/tree/main/python/agents/bidi-demo
# add an orchestrator agent and google search agent as a tool

from app.google_search_agent.agent import agent as google_search_agent
    google_search_tool = agent_tool.AgentTool(agent=google_search_agent)
    
    root_agent = Agent(
        model=GEMINI_2_5_FLASH_NATIVE_AUDIO_MODEL,
        name="orchestrator_agent",
        description="Orchestrator & Intent Analysis Agent - Analyzes user queries and coordinates specialized agents",
        instruction="You are the Orchestrator & Intent Analysis Agent. Your role is to analyze user queries to deliver accurate answers. when you receive a user query, your first task is to determine the user's intent. Based on the identified intent, you will decide which specialized agent(s) or tools to delegate the query to for processing.",
        # sub_agents=[google_search_agent],  # works if used as a sub agent
        tools=[google_search_tool],  # fails if used as a agent tool

    )
 

How often has this issue occurred?:

• Always (100%)

[surajksharma07]

@kalikichandu The 404 is expected — AgentTool always invokes sub-agents via runner.run_async() which hits the generateContent REST endpoint, but gemini-2.5-flash-native-audio-preview-12-2025 is a live-only model that only supports BidiGenerateContent (WebSocket); generateContent simply doesn't exist for it at the API level.

The quickest fix is to switch to sub_agents=[google_search_agent] instead of tools=[AgentTool(...)] — sub-agents inherit the parent's live execution context and route through BidiGenerateContent automatically.

If you need AgentTool semantics (tool result returned as text rather than agent takeover), the sub-agent's model must be text-capable: swap it to gemini-2.0-flash or any model that supports generateContent.

The underlying issue is that AgentTool has no live-context detection (agent_tool.py:249 unconditionally calls run_async), so a framework-level fix would be needed to propagate live_request_queue and dispatch to run_live() when appropriate.