I have been experiencing the same issue for several months: OpenHands is working very poorly with SOTA Chinese models. Recently, I tested it with Kim k2.6, and the issue seems to be with the agentic workflow. It works very slowly and gets stuck at many places. Especially I noticed this when we give a prompt and it starts working but if we stop the response and then tell it to continue, many times it doesn't resume. If it does start, it doesn't actually do any work and just keeps getting stuck at certain steps. It is hard to explain the exact issue properly in text, but I request all the maintainers to please try using any Chinese model (for example, Kim k2.6) in the cloud-hosted OH "All Hands AI". Just give it a medium-hard prompt and test it for half an hour. I promise you will see the issue for yourself; otherwise, it is hard to fully understand it. Please take this seriously. As the biggest open-source project in our niche, we should support open-source models better than any other platform. Arguably, this is the biggest weakness of OpenHands right now, but I hope it will be fixed soon. I request @wooden-umbrella-84098 to ask the maintainers about this major issue and take charge of it. I reported this earlier as well, but nothing has changed yet.

Hey @brash-engine-38895 we use Minimax pretty frequently, and it’s good but it definitely gets stuck more often. We’re just about to ship flexible model choice that will make it easier to switch between models, and when we do that I’m thinking of changing my default to an open model and then just bumping to the closed models for harder tasks

@limited-student-10747 But issue comes with the Kim k2.6 which is currrent opensource SOTA and also for deepseek v4 it have same issue, so i request you to pls do some major work to fix this, i am more a vibecoder 😅 but have a very solid experience in vibecoding complex projects, so let me know if i can help in this.

Yeah, basically I wanted to say we’re starting to work on it!

Great, if i can help in this then pls let me know

It may take a little while. BTW: have you tried deepseek-4 flash? https://openhands-ai.slack.com/archives/C06PB3T5ZK6/p1778111450905579?thread_ts=1778072285.841769&cid=C06PB3T5ZK6