← sources

The frontier of AI has stalled. Local LLM will too.

source: reddit (reddit:LocalLLaMA)published 2026-11-18 17:49ingested 2026-11-19 02:39

retained text

I’m fully prepared to be wrong about this, but I expect very few leaps in capability in local LLM.  I don’t mean there isn’t more to be gained with speed and size, but actual capability has hit a local plateau.  The difference between a 500B model and 9B model has shrunk and will continue to.  
Qwen4 will be a small disappointment.  The post training is taking much longer because they can’t squeeze any juice out of the Claude and OpenAI models because those models haven’t improved in ways that most people can see. Enterprise customers are paying through the nose for tokens and getting fed up quickly and moving to local Chinese models because of cost.  DeepSeek is getting overwhelmed and is basically out of compute it seems.    
I am not really upset because flash next is actually good enough for me most of the time and Claude subscription is still cheap enough.    
I hope everyone gets their small model that is good enough for them, but I don’t think the massive home LLM will be heads and shoulders over them like they were one year ago.  
Source: I’m an engineer working with LLM 

document fetched 2026-11-19 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1xzlq7g/the_frontier_of_ai_has_stalled_local_llm_will_too