#Would love to hear the final review -

1 messages · Page 1 of 1 (latest)

mental parrot
#

yeah that's my general feeling. There was a brief moment where the opensource models were much closer than they are now (probably around sonnet 3.7ish time). At that time, we spun up QWEN's massive coder model (and some others)... forgotten the code now... 480B? Can't remember... It was 95% of the performance as the best closed source options, but we could run it on our own servers and it reduced out monthly spend by tens of thousands.

But since then, the performance of the American models has advanced. Chinese models continue to benchmark well, but they're just not as performant in real world applications.

But it remains fun to just test these things!

mental parrot
#

Still going with minimax. Honestly it's pretty good! The $10/mo plan is sufficient. Though today was relatively light usage of clawdbot I got no where near the 100 prompt/5hr cap.

Provided you're not asking it to code it's perfectly sufficient.