little sidenote, guess Cursor still has that little annoying “feature” that you cant run your own oAI endpoint with reuglar cursor, you always gotta “untick” the “enable openAI api key” else the regular cursor models dont work 
maybe needs a feature request?
if you use Mac, there are keybinding for enable disable Key API. its Command+ Shift + 0
where do u see this setting?
The agentive capabilities of this model are truly incredible. I agree with you, it would really be a great choice, especially since it’s so cheap. Look, I use my own API key and I must admit I’m extremely satisfied, both with Psyche 3.2 species and with GLM 4.7, and now with the recent new MEmax 2.1. I don’t know if you’ve seen it yet, but if I had to choose today between MEmax 2.1 and LM 4.7, I would definitely go for MEmax 2.1. It has insanely impressive abilities, and since I’m using it in the course, I’m sure it will work well. The reason is that it hasn’t had a single tool output error since I started using it, and I’ve already used over 1,000,000 tokens with it in the course. GLM 4.7, on the other hand, does throw some tool output errors because of its output.
Minimax for Daily flow, GLM for Planning and BUg hunting, its also a tad better and understanding concepts and bigger picture
dont choose between them, use both where they shine
agree, i use MiniMax for Planning and Execution to make my daily limit more. sometime for execution using Composer or Gemini Flash for cheap or even using grok code fast
does it actually work for you?
Got the same key binding (though for windows) and it doesn’t toggle it at all 
Yeah, I’ve also been using MiniMax m2.1 and GLM 4.7 a lot at work recently. For their price points, the performance they deliver is pretty solid. MiniMax m2.1 is a bit stronger though😊
Hey everyone, long time no see! I’ve noticed something recently: GLM 4.7 performs totally differently across different tools. It shows amazing capabilities in Claude Code, but seems pretty average in Cursor, Opencode, Kilo Code, etc. Anyone else run into the same thing? In Claude Code, GLM 4.7 keeps insisting it’s actually Claude 4.5 Sonnet/Opus.
Been using GLM 4.7 with Zed via Openrouter. It’s not cheap man, but works. GLM is not the dream model you guys are all dreaming of. I’ve seen it hallucinate stuff, when I had to correct it multiple times - even though my context usage was 10%.
It repeatedly goes off-trail without even considering whats written in the AGENTS.md file - so I had to put a custom rule in Zed to “not create md files” and “look into the AGENTS.md file first before looking elsewhere”. My costs have gone done since, so do make sure you are using a proper ruleset for GLM as its a horse that needs some good directions.
Short answer is ‘no’. 4.6V accepts images, although haven’t tested that in Cursor.
its because of some system prompt but doesnt really matter, the neweest claude code version also tracks and shows which custom model you used (glm 4.5 air for resarch is often 30% in my list)
i can confirm that using 4.6V does work in cursor with images (last tried 2.5 weeks ago)
but glm is just too slow in cursor
it works really well in trae.al, but i guess its because both are based in singapore and the speed sint much of an issue
GLM-5 has been released. Any plans to add it? This should help us use more tokens.
Following, @condor any updates on this?
Hi @hmeng I do not have a timeline for it yet. We recently added Kimi K2.5 to Cusror and each model takes some time to properly test.
