Provide Letta LLM Leaderboards

[distributev]

So that people will clearly see which LLMs work well with Letta and which do not.

Similar idea with

https://aider.chat/docs/leaderboards/

Meanwhile 'Berkeley Function-Calling Leaderboard' is good enough because it ranks LLMs based
on their ability to do Function-Calling (which is fundamental for Letta's well functioning)

https://gorilla.cs.berkeley.edu/leaderboard.html

Here you can see that GPT-4o-mini-2024-07-18 (FC) is a great choice for Letta because it ranks 4th and
it has a very good price.

P.S - I lost a huge amount of time trying MemGPT/Letta with lots or random LLMs,
including the "stars" LLMs, just to get back stack traces up to the level of almost giving up.

Until, by luck, I tried GPT-4o-mini and it worked great + when I saw the price I knew this is it.

[sarahwooders]

We just started working on this recently, so will have updates toon. Anecdotally, we've also seen gpt-4o-mini (but not gpt-4o) have great cost/performance generally recommend that.

[distributev]

Another good idea and not difficult to put in practice is again what aider is doing (even if they have the leaderboard). The 3rd sentence on their home page is

'Aider works best with GPT-4o & Claude 3.5 Sonnet and can connect to almost any LLM.'

They tell people upfront what works well and then, with the leaderboards, they provide more details for people
which need more details.

[github-actions]

This issue is stale because it has been open for 30 days with no activity.