Do open models just need more parameters to catch up?
The poster argues that Anthropic and OpenAI don't have any special secret technique — their real advantage is simply model scale, meaning parameter count. According to rumors, Opus has around 5 trillion (5T) parameters and the models around 10 trillion (10T), while stayed under 1 trillion (1T) parameters for a long time.
Only recently did DeepSeek V4 and Kimi K3 break past that 1T ceiling, and jumped noticeably as parameter size grew. The core claim: the gap between open and comes down to raw scale, not hidden algorithmic tricks.
Key points
- Claim: Anthropic/OpenAI's edge is scale (parameter count), not secret techniques
- Rumored parameters: Opus ~5T, models ~10T
- stayed under 1T parameters for a long time
- DeepSeek V4 and Kimi K3 recently broke the 1T ceiling with a notable jump
- All parameter figures are unconfirmed rumors