Reddit debate: Anthropic & OpenAI's moat is just scale, not secret sauce

A thought-provoking Reddit discussion has ignited debate about the true source of advantage for AI leaders like Anthropic and OpenAI. The post claims these frontier labs lack a 'secret sauce' and instead depend entirely on massive scale. Rumored parameter counts — 5 trillion for Opus and 10 trillion for Mythos/Fable — dwarf the sub-1 trillion sizes typical of open models until recently. The ceiling was only broken when DeepSeek V4 and Kimi K3 pushed past 1T parameters, reportedly yielding noticeable performance gains.

While unconfirmed, the speculation underscores a growing tension in AI: is the gap between closed and open models narrowing purely because of hardware and data scale, or are there architectural innovations we don't know about? The post suggests that if open models can match scale, the 'moat' of frontier labs evaporates. For professionals, this raises strategic questions about where to invest — in proprietary APIs or in open weights that may catch up as scale becomes commoditized.