What local models do you use? #112
Answered
by
mlhher
junglebells
asked this question in
Q&A
|
Open to all, but specifically for @mlhher since you mentioned that late is built using late. What models do you use for primary and sub agents? |
Answered by
mlhher
Aug 25, 2026
Replies: 1 comment
|
I used to primarily use Qwen3.5/3.6 35B-A3B for anything coding Gemma-4-26B-A4B for anything that is not just "pure" coding. Both K_XL quants Q8 though Gemma seems noticeably more affected by the cached quant than Qwen. Recently I have been using DS-V4-Flash-0731 as it is quite exceptional in both speed and performance and seems to do really well with the subagent architecture. And it is extremely cheap (300 mio tokens total, 2.8$ used all from late). The recent price hikes should not have changed this a lot. |
0 replies
Answer selected by
junglebells
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I used to primarily use Qwen3.5/3.6 35B-A3B for anything coding Gemma-4-26B-A4B for anything that is not just "pure" coding. Both K_XL quants Q8 though Gemma seems noticeably more affected by the cached quant than Qwen.
Recently I have been using DS-V4-Flash-0731 as it is quite exceptional in both speed and performance and seems to do really well with the subagent architecture. And it is extremely cheap (300 mio tokens total, 2.8$ used all from late). The recent price hikes should not have changed this a lot.