Deep Research Only Looks for Sources with OpenAI

[degan6]

Hello!
I've been testing out gpt-researcher with the default OpenAI setup and have been impressed with the output. But when I switched to using Ollama with the GPT-120B model, the output quality dropped quite a bit.

I know that the models aren't directly comparable, but the issue seems to be in the research steps. Where I see the default OpenAI config make as many as 30 searches with Tavily, all other providers I've tested, including Anthropic, make between 0 and 1 search.

I'm not sure what could be causing the issue. Is this expected behavior?

Here is my config

TAVILY_API_KEY=tvly-<key>
SCRAPER="tavily_extract"
RETRIEVER=tavily,mcp

ANTHROPIC_API_KEY=sk-<key>

FAST_LLM='anthropic:claude-sonnet-4-5-20250929'
SMART_LLM='anthropic:claude-sonnet-4-5-20250929'
STRATEGIC_LLM='anthropic:claude-sonnet-4-5-20250929'

I'm running something like:

python cli.py "What is happening in Portland this month in 2025" --report_type deep

Once the model starts, it immediately starts writing the output. When I run the tool with OpenAI models using the OpenAI API, it spends 2-3 minutes researching sources and going through the deep research steps.