That explains a lot. I’m glad many people had a positive experience, but… I tested it again! And again!
I got terrible results.
Did anyone on your team get better results than I did?
Especially regarding compliance, following the rules, and sticking to instructions?
I find it hard to follow — who do you actually call a “partner”?
A true business partner? Or even a company you work with, like OpenAI, is also called a partner?
Did you check my complaints?
Did it follow the rules?
Did it stick to the instructions without deviating?
Right now, only GPT-5 is good at that.
@yakovw A partner can be anyone you consider a partner. Not just a business or life partner but also someone you work with or collaborate with on a project. In this case, yes it is about providing test access to users to explore a model and its capabilities.
Can you tell me if your team got a different impression than I did?
I think this could cause them embarrassment…
Is this post shared with them, and will they see it?
After all, that’s supposed to be the purpose of the experiment.
It would be better for them to fix or improve it first.
Nothing to worry about (the admin note and reveal). We merged a few threads to make it easier for people to find the discussion.
It’s a pity that the rates are hidden. It is impossible to estimate the quality/cost.
Sonic is about as dumb as GPT-5 Mini: both were given a task to make a change to a PowerShell script; looked at its output; didn’t realize they hadn’t completed the task; and reported that the task had been completed.
I gave up and did it through GPT-5-low.
For simple tasks it is worth trying to run Sonic, but you need to check what it did. It can lie in the reports.
The Sonic model called itself Grok’s brother tuned for coding, and I have to admit it’s not that bad. I think it’s somewhere around the Auto level. I gave it a few simple tasks and it did them without any problem, but it struggles more with the complex ones.
Did you do some kind of manipulation?
Did you ask it to write that?
Because I can’t manage to reproduce it at all.
Although unfortunately, I can’t think of another company that could create such a successful code model.
nothing manipulation at all
Interesting why I can’t manage to get the same result.
You convinced me.
In fact, there’s no need to convince me anymore.
I have a specific question that only Grok answers,
and now I checked—
indeed, it’s one hundred percent him.
You can guess what I mean if you’d like.
Good luck, and thank you for taking the time.
When you ask it, it confidently says xAI, but it feels very Anthropic
The processing speed is very fast, but the ability to handle complex projects is insufficient. It may be necessary to reconstruct the project from the beginning and use standardized documentation to demonstrate its capabilities.
