1–10 of 11 posts
Which model are you using now? Why? What are the good and bad parts?
cheap, fast, not as good, but more gets done without giving Big Ai any more money
I use qwen3.6:36B locally
other peoples compute resources are unreliable ar best.
deepseek v4 pro - but honestly it is comparable to gpt-5.4-mini, far from GPT5.4 let alone GPT5.5! Its advantage is really just its pricing. I'm going to give Kimi K2.7 a try - with K2.6, even its cloud chat locks up all the time, so it really didn't give me much confidence at all for agentic coding.
deepseek v4 pro - but honestly it is comparable to gpt-5.4-mini, far from GPT5.4 let alone GPT5.5! Its advantage is really just its pricing. I'm going to give Kimi K2.7 a try - with K2.6, even its cloud chat locks up all the time, so it really didn't give me much confidence at all for agentic coding.
I think it is more like a post-training issue. DeepSeek is cheap enough for heavy token use, so it fits Hermes very well.
Look at the gap between gpt5.4-mini vs deepseek v4 pro!
The main reason I'm using Chinese LLM is cost. Minimax M3 is a good deal with large context window. But, M3 jumps into implementation too quickly even when a task is clearly defined. It misses tests or edge cases, and occasionally lose track during longer work.
i built in skills to work with M3 in a service called typed, an ai cli. it uses m3 under the hood (up to ~500k tokens), then switches to deepseek for up to 1M. a few bells and whistles added of typescript/python coding optimization. and just built a custom TUI frontend for it (initially works with the claude code tui and still does).
to toggle the typed tui you can run:
typed cli on
typed cli off