DeepSeek open-sourced DeepInfer, the engine behind its API.
Its receipt: $38 of revenue per GPU-hour (off-peak). In 2025 the same math gave ~$13 on H800s. The catch: 64 + 64 Blackwell-class GPUs, and no cache store in the repo.
DeepSeek DeepInfer is the engine behind DeepSeek's API, open-sourced Oct 20, 2026.
DeepSeek says one GPU earns $38/hour off-peak. The tricks: split prefill/decode, decode across 64 GPUs, 96% cache hits (reported). The catch: the cache isn't in the repo.
HY4 vs DeepSeek V4.1 Flash 1017
My first thoughts. HY4 is a fucking expensive overthinker. Painfully slow, thinks way too much, and the final quality is only slightly better than DeepSeek. and guess what? this single test cost me $31.41 via AIHubMix 💀 I'd pick DeepSeek
Why every open-weight AI model will ship as a torrent 🧲
Torrent networks suit model weights better than anything else shared on them, e.g. movies. Model files are large and everyone wants the identical copy. A torrent is named by a hash of its files. DeepSeek-V4.1-Flash-1017
You have reached the end of the archive
All of deepseek