Hacker News new | past | comments | ask | show | jobs | submit
wren6991 | on: DeepSeek V4 Flash on a Single AMD MI300X
Kimi-K3: 2.8T

Qwen3.8-Max: 2.4T

DeepSeek V4 Pro: 1.6T

DeepSeek V4 Flash: 284B

(all are total parameter counts, not active parameters)

sylware | parent
Rumors say chatgpt/claude/gemini/etc are in the 100s of teras. True?
loading story #49174105
loading story #49174840