thin birch
In chatGPT.com web-ui, When I ask a deep-research question, I can choose any model from the model-picker. GPT-4o, GPT.4.5, o1, o3-mini, o3-min-high, all are vaialable (at this time of writing).
Which one should I pick, or is this arbitrary because there is default model that will be selected anayway, regardless of the current choice of model-picker items?
dense knotBOT
Hi @thin birch. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
When deciding which model to pick from the model-picker for deep-research questions in ChatGPT, it's important to consider the specific strengths and capabilities of each model. Here's a brief guide based on the available information:
GPT-4o: This is OpenAI's flagship model that can reason across audio, vision, and text in real time. It's suitable for tasks that require a high level of intelligence and the ability to handle complex inputs, including text and images. If your deep-research question involves multimedia inputs or requires advanced reasoning, GPT-4o might be the best choice. Source.
GPT-4.5: This model is a research preview and is noted for its improved ability to recognize patterns, draw connections, and generate creative insights without reasoning. It might be useful for tasks that require a broader knowledge base and improved user intent following. Source.
o1 and o3-mini: These are reasoning models designed for complex problem-solving tasks. The o3-mini, in particular, is optimized for coding, math, and science, and offers faster responses with cost efficiency. If your deep-research question involves advanced reasoning or coding, these models could be appropriate. Source.
o3-mini-high: This model is a variant of the o3-mini, likely offering enhanced capabilities for reasoning tasks. It would be suitable for more intensive reasoning tasks where speed and depth are required.
In general, the choice of model should align with the nature of your deep-research question. If the task involves straightforward execution and speed is a priority, GPT models like GPT-4o might be more suitable. For tasks requiring complex reasoning and decision-making, the o-series models are recommended.
thin birch
@dense knot If I choose a resoning model for deep research, will all intermediate steps (exploring - gathering- summarizing - whatever else) be prepended with a reasoning step, or just the first one (making a research plan, drafting a outline...), and then the execution of the plan is done by a nonreasoning model?
dense knotBOT
Let me check the knowledge base for you again. One moment! :books:
When you choose a reasoning model for deep research, the reasoning model is primarily used for tasks that require complex problem-solving, strategizing, and planning. According to the information available, reasoning models like the o-series are particularly effective at multi-step agentic planning. This means they are well-suited for creating a detailed, multi-step solution to a problem, such as making a research plan or drafting an outline.
Once the reasoning model has developed a plan, the execution of the plan can involve both reasoning and non-reasoning models, depending on the specific needs of each step. The reasoning model can orchestrate the workflow, selecting and assigning the appropriate model for each task based on whether high intelligence or low latency is more important. For example, the reasoning model might handle the initial planning and complex decision-making, while a GPT model could be used for executing specific tasks that require speed and efficiency.
In essence, the reasoning model acts as "the planner," and it can delegate execution tasks to "the doer" (a GPT model) when appropriate. This approach allows for a combination of strategic planning and efficient task execution, leveraging the strengths of both model types.
thin birch
@dense knot for deep-research questions, for customers on the "openai plus" tier, is the activity contrained to a specific maximum duration of say 20 minutes +-5 mins per question? is it unconstrained (or less contrained) for customers on the "openai pro" tier?
dense knotBOT
Let me check the knowledge base for you again. One moment! :books:
For deep-research questions, the duration of each research task is typically between 5 to 30 minutes, depending on the complexity of the query. This applies to all users, including those on the "OpenAI Plus" tier. The process involves the model autonomously searching for and reading information from diverse online sources to create a thorough, documented report.
Regarding constraints based on the subscription tier, the information provided does not specify different time constraints for "OpenAI Plus" versus "OpenAI Pro" users. However, "OpenAI Pro" users do have a higher limit on the number of deep research queries they can perform per month—120 queries compared to 10 for Plus users. This suggests that while the duration per query might not differ significantly between tiers, the volume of queries allowed is greater for Pro users.