Are you at least able to confirm if the model provider supports the legacy privacy settings?
I do not see DeepSeek here and hence we have block listed them, but I have no clue about supernova’s respect for our privacy settings
Have you made it work with MCP? is doesn’t seems to work
On my side it works
Or you could have told it to not focus on the container of water but why it is outputting twice as much as expected at two locations instead of one. Identifying bugs is honestly one of main uses of Cursor. People who aren’t programmers, don’t know how to describe the issue and give it proper context. “Cup is overflowing, fix!” would obviously have the model focused on the container not the source issue. For all it knows, the leaking is a feature you want.
1 Like
It’s also important to choose the right one (smart enough?) model and watch out for context overflow. I recently fixed a flaw through an Agent (Gemini 2.5 Pro) and he handled it more correctly and concisely in a different place of the program than where I expected to rewrite a bunch of code.
It is technically capable but programs things a human wouldn’t program, like “Signup now completes successfully even if confirmation email fails to send”
1 Like
WAY too fast to be Sonnet. Unless Anthropic has invested in a “Collosus” sized mega-datacenter with top of the line hardware, its highly doubtful this is Sonnet. I love sonnet, but the thing crawls around like a slug.
Unable to access this model
Admin settings
Additionally, max mode is disabled across the teams
for those curious, may be this might help guess the model
I’ll be disappointed if it’s really Grok 4 Fast, because it’s dumber and slower than Grok Coder Fast… while the delicious Grok 4 is still ■■■■■■■ broken.
no.
Nobody said if it’s actually good? Better/faster than grok-code? Level of intelligence? Ignore, or worth a look?
Codex is ripping it for me, just need a fast “good enough” companion model, to unsub from Claude code.
It’s.. ok.. it’s good enough for most use cases but it has a hard time figuring out some complex integrations i’m doing with rust/c++ and other apps even with their source code in the workspace. But it’s really good at reading everything and making documentation and ploughing through cleanup and things like that, i just don’t think it can figure out complex issues. But other models can do that like GPT5 or one of the others if you get stuck then go back to this one. I feel like it’s similar to grok-code-fast-1 but different.
Best thing you can do is give it a shot since it’s free make it do a lot of work and test it. Clone the workspace into another workspace and let it rip
1 Like
I tried it for one day and it was substantially worse the Claude and even Gemini
2 Likes
Thanks guys, valuable feedback!
probably worse than grok-code-fast-1 or whatever that model is called, i told supernova to do depth-first-search and it did breadth-first-search instead, which is the exact opposite. it did fix itself quite well when pointing out problems with its implementation, however all i would praise is its speed.
1 Like
grok-code-fast-1 seems to be better overall. Code-supernova does follow instructions regarding planning, grok just likes to jump in and start making changes.
I’ll be still testing, but grok-code is still my daily driver for now.
2 Likes
Also my very first impression - both rather dumb, and grok-code faster, thus winner.
on the scale of 0-10 rips, how rippy is codex vs cursor agent?
1 Like
It’s about the model, not the tool. gpt5-codex-high would also rip in Cursor (once it’s adapted, and gets enough context which gpt5 seems to).
Codex rips it out of everybody else, another world. I lost trust in Sonnet 4 (Claude code), since codex (and gpt5) are so much better. It takes a while tho, different workflow, big complex tasks in parallel.
3 Likes
codex with gpt-5 
cursor with gpt-5 
2 Likes