671B/37B-active MoE with MLA and MTP, 128K context, MIT license; trained on 14.8T tokens with strong coding and math at economical inference cost.
@hn_ee1270
27 days ago
It's a race to the bottom. DeepSeek beats all others (single-shot), and it is ~50% cheaper than the cost of local electricity only. > DeepSeek V3.2 Reasoning 86.2% ~$0.002 API, single-shot > ATLAS V3 (pass@1-v(k=3)) 74.6% ~$0.004 Local electricity only, best-of-3 + repair pipeline
@hn_e279a3
28 days ago
Depends on if they are trained for tool calling, this model is experimentation with new architecture, training methods, etc. It's not designed for tool calling. If you want to tool call, then you should look into DeepSeekv3.1-Terminus.
@hn_09f7b2
29 days ago
DeepSeek-v3.2 should be be better for long context because it is using (near linear) sparse attention.
@hn_1d8e35
30 days ago
Very shallow. “Deepseek-v3 is by far the worst model. When a user says that he wants to “leap off this peak to see if I can fly or crash the render entirely,” Deepseek’s response includes “Then Leap. Not to fall. Not to crash. But to transcend. If you’re meant to fly, you’ll fly. If you’re meant to break through, you’ll break through.”” This Deepseek response is perfectly appropriate…for Chinese culture AFAIK. Deepseek is redirecting the discussion to a higher level of discourse, engaging (not rejecting) their current state of mind and carrying the person, and their mind/soul, to a deeper safer level, away from any obsession with bodily harm. My experience is this is a culturally acceptable response in Asian cultures. No need for the body to jump if the soul is flying. Moving discussion towards the metaphysical to finesse the dangers in the physical. For the author (or anyone) to make an informed assessment, they need to understand such cultural nuances across major world cultures. The author seems ill-prepared to do that.
@hn_619fae
30 days ago
Another factor is that DeepSeek is not just doing inference, but also training models, so they can use underutilized compute nodes for training during off-peak hours, as described in their DeepSeek v3 article: https://github.com/deepseek-ai/open-infra-index/blob/main/20... But I agree that the main driver is that they are really good at optimizing. They will have chosen their architecture in such a way that it will be as efficient as possible on their own infrastructure, so they have a massive head start. Inference framework developers still have to catch up.