I was one of the first people to try K2 on Groq via OpenRouter last night, and everything was running smoothly. But this morning, we’ve been running into a lot of issues with tool calling.
As you can see from the screenshot below, tons of API calls are being wasted due to failed tool calls. The ones that finish with "tool_calls" are the ones that work — all the others are failing.
Just checking if you’re aware of this and if a fix is on the way. Cheers!
Hi there, I’ve reported this to the Kimi team; we’re adding a ton of optimizations right now, will get it fixed ASAP
Hi, is there an update on this issue?
We’ve rolled out a few fixes which should have made it better; there’s another rollout going out soon. Some of this has to do with how the API processes tool calls. Any failed tool calls shouldn’t be charged to your account though.
We’ve rolled out a few fixes which should have made it better; there’s another rollout going out soon. Some of this has to do with how the API processes tool calls. Any failed tool calls shouldn’t be charged to your account though.
Hi, I’m still getting errors or responses like function call get_webpage_content when, for example, I try to make a web search and have the model (Kimi K2) browse the results to create the answer.
Yes, things are much better but I still face them like 5% of time when using OpenCode
Thanks for reporting, we’re looking into these now
