By somebody |

We offer several different models for search.  They vary in cost, and vary even more depending upon the number of documents you ask to be returned in your queries:

model cost table

The above table shows how many queries (approximate) you can make on a $10 budget with an average context document limit of 10.

Of course, the Gold and Silver models are always going to give you the very best responses.  But as you can see from the table, you can ask a heck of a lot more questions to the less expensive models for the same cost.  So, how do you use these cheaper models for research?

As it turns out, these models are very good when it comes to hard facts: who, what, where when -- specific persons, places and things.  They are far less reliable when it comes to higher level queries where we are looking for concepts and ideals.

How do you use the less expensive models for researching hard facts?  Use keywords whenever you can.

Examples:

Who fell from the loft?

This is a Gold level response:

who fell from the loft - claude v3

Compare the above with these Bronze level responses, using keywords to help narrow down the searches:

who fell from the loft - gemini prowho fell from the loft - gpt-3.5-turbo-16k

What did God say to Michael regarding the men who faced East away from the His temple?

what did god say to michael - gemini prowhat did god say to michael - gpt-3.5-turbo-16k

As you can see, the responses from gpt-3.5-turbo-16k are consistently better than those of gemini-pro, but gpt-3.5-turbo-16k has a very small context window (only 16k) compared to gemini-pro (32k).

Hopefully you can see that these less expensive "Bronze" models are quite good when it comes to straight facts when given enough detail.

However, when it comes to more broad, generalized thoughts, concepts, ideals -- forget about it.

What was the motivation for the Israelites to build a golden calf?

what was the motivation for the golden calf - claude v3

gemini-pro can't answer this at all.  gpt-3.5-turbo-16k's response is woefully inadequate.

what was the motivation for the golden calf - gpt-3.5-turbo-16k

In conclusion, when you're trying to find the names of people who did stuff, things that happened, places where they happened -- the Bronze models should do just fine (but you may need to add keywords).  When you need to ask the bigger questions that require more introspection and depth, go for the Gold.