Like descristion, i want glm-4.6 be added to Cursor. I’m still waiting for this model in Cursor (i means at least 4.5) but till now, it’s not. What wrong ?
18 Likes
I am using it. Add your api and endpoint, then add the custom model glm-4.6
1 Like
That breaks openAI models, some people want to be able to use both.
2 Likes
Cursor team should try creating different context for tool calling and instruction for different models
This will solve the issue there are no more that 13 model of which an individual only uses at most 5
So how hard it is to create a custom instruction specific to those 5 models
1 Like
Seems Claude-4.5-sonnet also disappears when using my zai api. They should simply separate their subscription models from personal api models
I asked for that feature, but it really didn’t get the track…
1 Like
We need an easy way to integrate new models when needed. Not predefined with OpenAPI, Gemini or Azure setups, should be something with Custom fields like Roocode or the others provides support for new models, easy interface name of the model, api route and key thats it
2 Likes
Please add GLM support for regions where Claude and GPT are not currently available.
2 Likes
how ?
GLM4.6 is pretty good and much better than GLM4.5. Hope can see it in cursor model list, otherwise I have to use claude code + GLM4.6.
2 Likes
This is never, ever, going to happen.
People have been asking for GLM 4.5 for months.
At best, you will have them telling you a lie that “they are looking into it” or “have notified the team” before they proceed to ignore you and the request for a couple of months, or they will claim it’s really, really hard to for them to do it.
Beats me why, but they just straight up do not want to give access to GLM models but also aren’t going to clearly say so.
3 Likes
GLM’s capabilities have already matched those of Claude, and if they were to add further enhancements, it would jeopardize their collaboration with Claude—clearly a move aimed at exclusivity.
1 Like
I think Cursor doesn’t want to rely too heavily on Chinese providers — it’s a major risk right now due to political factors.
It also allows Cursor to manage bulk limits more effectively.
If they purchase Claude tokens but can’t use them because of GLM, that would create a lot of complications.
Exclusivity might actually matter here, as it allows Cursor to secure better wholesale pricing deals.
This proves the model’s strong capabilities, so I’m planning to shift some of the funding to GLM.
Yes, that’s the expected behavior in this situation. However, I have to admit that I already tried switching to Claude Code with its generous limits and was disappointed. The efficiency and autonomy of Sonnet 4 and Opus 4.1 were far lower than in Cursor — even when using Roo Code, which solves many issues.
As far as I know, Z Code lags significantly behind Claude Code in quality.
Still, I recommend using them through Roo Code to mitigate these issues.
1 Like
Cline, roocode, and kilocode are all fantastic.
1 Like
++i need this model too
IDK, I gave GLM 4.6 via Roo a simple task yesterday, asking it to change epoch millis to a date/timestamp with local time zone in the UI. It failed. I told it it was still epoch millis. It failed again. Both times it pronounced success. Then, I handed it to Cursor Auto, which fixed it, saying the root cause was
The API is returning epoch milliseconds as strings (like “1758320470264”), but the date formatting functions were only handling ISO date strings and other standard formats.
GLM created the date formatting functions Auto was referring to. Why did it assume ISO date when I told it it was epoch millis?
Obviously, the success of GLM 4.6 is only good in comparison with its price. Claude 4 and GPT 5 still have a significant advantage.
2 Likes
yes as long provide to fireworks