Multi-agent which model are you guys using as a team agent?

General
1

Hi there, regarding multi-agent systems and the best LLMs for reasoning: I have a multi-agent setup with a team agent and four task agents. In my tests, Claude outperforms all models. O3-mini is quite good but too slow. However, recently, I’ve been getting too many 529 - overloaded_error issues with Claude, so I’m testing a reasoning agent using GPT-4o with DeepSeek.

  • Question 1: Which model are you guys using as a team agent?
  • Question 2: Regarding reasoning, what would be the best combination?

Important notice: One of these agents is responsible for retrieving a list of ~200 clients to obtain client_id values for use in other agents. Some models, like Qwen, cannot handle this and return the following error: “Limit: 6000. Requested: 11,620. Please reduce your message size and try again.” Thanks in advance

2

Hi @edudesouza
Thanks for reaching out and for using Agno! I’ve looped in the right engineers to help with your question. We usually respond within 48 hours, but if this is urgent, just let us know, and we’ll do our best to prioritize it.
Appreciate your patience—we’ll get back to you soon! :smile:

3

Hi @edudesouza

  1. The team agent doesn’t typically have to be a very good model, since it only has to perform basic function calling and collation of responses. A gpt-4o level model should be sufficient.
  2. For reasoning You can use deepseek-r1-distill-llama-70b via Groq. That is pretty good. Otherwise the o3 models from OpenAI. To make it perform better, set the reasoning_model of the agent that has to reason, and make the main model something normal like gpt-4o, then the reasoning is a separate step that would only reason and then a faster model can do the rest of the work.

For your actual problem, I can suggest not giving the agent too big a task to handle, I can imagine that context window becoming massive, especially if history is enabled. If you can use automation around the agent to simplify the problem space, that would be recommended.

4

Could you give me an example whether to use reasoning? Since we have models like 01-mini that has his capability

5

So yes you can just use o1-mini and provide reasoning_effort which indicates to the model to try to reason. If it responds with reasoning content, we use it.

So you could do something like

from agno.agent import Agent
from agno.models.openai import OpenAIChat
from agno.tools.yfinance import YFinanceTools

agent = Agent(
    model=OpenAIChat(id="o3-mini", reasoning_effort="high"),
    tools=[YFinanceTools(enable_all=True)],
    show_tool_calls=True,
    markdown=True,
    debug_mode=True
)

# Print the response in the terminal
agent.print_response("Write a report on the NVDA, is it a good buy?", stream=True)