Testing thinking levels in DeepSeek V4-Flash 0731’s
Same neon Tokyo prompt at off, low, high and max on 2 DGX Sparks
Progress on the agentic newsroom.
Built in my own harness with my daily drivers GLM 5.2 + DeepSeek V4 Flash. It's creating bespoke reports based on what people are actually talking about. It has one job, and uses many tools to do that job in real time. The intent and goal is
DeepSeek is now able to read pictures with your eyes open, and comes with DSH’s simplest upgrade and usage instructions.
One picture, one sentence, 4 cents for me to develop a CS Counter-Strike 1.6.
At present, the price of DeepSeek-V4-Flash-Vision-Exp multi-modal model is acceptable, and it can be connected to its own DeepSeek Harness. Cache hit 99% Cache output speed 111 tok/s
You have reached the end of the archive
All of deepseek