Jev holds the white stone in the back hand. Basically, I use the DeepSeek 4.1 flash model to calculate the move, and it comes out in seconds...
Of course, in terms of results, whether it is poker or backgammon, the Agent with a large language model will usually win...
I've spent 2 days in testing Deepseek V4.1 Flash in game development capabilities using Godot Engine V4.7.1
It uses a total of 660M tokens. Video I record here shows the complete 2D side-scrolling platformer game that it managed to develop via DS Harness, I did not open the
DeepSeek-V4.1-Flash
The same prompt you used for the DeepSeek-V4.1-Flash as well as the Grok 4.6 . Everything you can see, I rendered and looked at first.
Cactus released Needle 3, an 8-29MB automation model matching DeepSeek V4 Flash.
One weight set, depths 2-20 layers, each its own model. Runs on Pi 5 at up to 4,000 tok/s. Not for chat, function calls only, returns empty if no tool fits.
You have reached the end of the archive
All of deepseek