Hi all, I'm curious about what open weight models are competitive right now based on your experience (everyone here )? I was using GLM-5.1 so far, but I'm hearing good things about Deepseek V4 Flash and it seems quite cheap compared to GLM-5.1.

https://index.openhands.dev/home You can filter by Openness

Thanks a lot for sharing! I definitely found this useful when I chose GLM-5.1 🙂. I didn't see DeepSeek V4 Flash in there yet. It was released a couple of weeks ago I believe.

I believe it is on our TODO list. @millions-crayon-32880 can confirm or correct me if I'm wrong 🙂

By the way I also read good things about Kimi 2.6 (this one seems to be in your table but the scores are missing 🙂 )

Yeah, right now we are trying to fill in our llm support tracker, and once it’s filled in we’ll consider adding new models

I can confirm that DeepSeek V4 Flash is working nicely on openhands after having tried it today. It's ~10 times cheaper (vs GLM 5.1) and seems to work quite well, plus it has a 1M token context window.

How can i use Deepseek v4 Flash or Pro

Sorry for the late reply, I actually just went to the LLM settings and added the openrouter base URL, and the model name as in open router (with the prefix "openrouter", e.g. openrouter/deepseek/deepSeek-v4-flash). By the way Kimi K3 also seems to work like a charm, though I decided to lower the thinking effort to low because it does seem to overthink a lot with thinking effort high. I guess it depends on the problem.