LocalLLaMA
DeepSeek v4 Pro is too big for such a "midrange" performance, or am I missing something?
Hi. DeepSeek v4 Pro has 1.6T parameters, probably the largest in open models, or at least one of the largest. Yet it's not the best/most performance open model, considering a wide variety of definitions of "best". Indeed, in most cases, it is not the second best, third best, or fourth best either. GLM 5.1 with 750B parameters is less than half the size of it, but is considered by many "an opus" in open models. So is Kimi K2.6, with 1T models, still far less than 1.T of...
Read the full discussion on Reddit
This article was aggregated from Reddit. Click to join the conversation.
View on Reddit