What is taking Deepseek so long to release a model ?
Posted by power97992@reddit | LocalLLaMA | View on Reddit | 14 comments
I hope they release a frontier model soon and it doesnt crash the market too much… better to release something than nothing!
Ok_Warning2146@reddit
They need to make it work with Huawei hardware which is quite buggy to begin with.
power97992@reddit (OP)
Zhipu trained glm 5 on 100k huawei gpus … I think they should have solved the huawei problem by now
Ok_Warning2146@reddit
Well, zhipu and DS are supposedly different teams.
power97992@reddit (OP)
Deepseek’s team should be more capable than even zhipu…
Ok_Warning2146@reddit
DS is better in research. That doesn't mean they are better in engineering.
Pleasant-Shallot-707@reddit
There was a sell off when they released R1 because it came from nowhere, was pretty good compared to frontiers, and was very cheap. They’re not repeating that
power97992@reddit (OP)
Man ds v3.2’s output tokens are almost 10x cheaper than kimi k2.6 but also way worse
nullmove@reddit
What do you think asking this here accomplishes?
power97992@reddit (OP)
Idk, maybe someone can give a reason for the delay…
PotatoeswithWifi@reddit
yeah waiting on training runs can feel like forever when you just want to try something new. i had the same issue with being stuck waiting on updates from one place. been using Modelsify for and it’s been a bit more flexible so far, gives access to different stuff without as much waiting in between
DepartmentOk9720@reddit
The right and only answer I can think of to these questions.
MuzafferMahi@reddit
I think they prioritize their scientific non-profit part more than other labs
Due-Memory-6957@reddit
Deepseek IS the scientific non-profit part.
Due-Memory-6957@reddit
I hope they realize whenever they want, but that it crashes the market a lot.