“DeepSeek V4.1 Flash is 98 percent as good as GPT-6 Astra.
“The cost is 1.4 percent.” What does this sentence remind you of? A comparison or a collision? No, no, it's nothing like that. We believe that costs can be reduced and better results can be achieved by dividing the work among these different models.
This time I had DeepSeek V4.1 Flash try the 5x5 Rubik’s Cube challenge...
It beat Astra by 2 seconds. Through computer use on a virtual cube as well. Not the same conditions as a human solving a physical cube, but still pretty wild to watch lol. I got to use DeepSeek V4.1
You have reached the end of the archive
All of deepseek