I just tested the deepseek.com API again at 11:21pm New York time, and now it's chugging along at 254 tokens per second. At that rate, the API is producing 1 million tokens in just over an hour. You can get lots of software written, and lots of tasks completed, at 1 million tokens per hour. At that rate you don't need a genius LLM to get work done. This hard working, inexpensive model has plenty of intelligence, and is able to gets jobs completed because it iterates relentlessly through planning, research, build, and test loops, at a speed other models can't match.

I just tested the deepseek.com API again at 11:21pm New York time, and now it's chugging along at 254 tokens per second. At that rate, the API is producing 1 million tokens in just over an hour. You can get lots of software written, and lots of tasks completed, at 1 million tokens per hour. At that rate you don't need a genius LLM to get work done. This hard working, inexpensive model gets jobs completed by iterating relentlessly through planning, research, build, and test loops. It has plenty of intelligence to do every piece of that looping work, to be constantly successful..

I just tested the deepseek.com API again at 11:21pm New York time, and now it's chugging along at 254 tokens per second. At that rate, the API is producing 1 million tokens in just over an hour. You can get lots of software written, and lots of tasks completed, at 1 million tokens per hour (even if you're not using the most genius LLM).

I just tested the deepseek.com API again, this time at 11:21pm, and I'm getting 254 tokens per second. Wow that's fast.