SSH

Yes, this was Qwen4 (my speculation is that this is Qwen4 plus, but it can be any variant) making a black hole in C++ with CUDA with much more compute given to get the best results. This is much more hardcore than my normal black hole bench test, and it did very well. When multiple frontier AI models review and grade it's result as done with all of the other Black Hole Benchmark 2.0 results, it is above Kimi k3.1 Pro and GPT 6.1 Terra overall.

Finally got to test Inkling 2 Small with all my Qwen 4 testing...
This is alright, but it is messy and has some issues. Not bad though.

Qwen 4 is cooking!

I built a feature in my T3Code fork where orchestrator agents can rate on a scale of 0-5 how well a subagent is doing for them
Currently, Qwen4 is rated higher than GPT 6.1 Sol

Also had the same model make a nothingburger, had the last qwen max model do this as well and it is MUCH better. Qwen is cooking!
It is a good orchestrator as well and knows properly when to call in other models for review