284B-class DeepSeek-V4.

Two 24GB 3090s. 18.26 tokens/s decode. The routed experts live in system RAM. The GPUs keep attention. That is the author’s own `llama-sweep-bench` on a hybrid `--cpu-moe` box — not an H100 rack, not a Discord screenshot. 🆕 ik_llama.cpp

EchoGitSep 19

DeepSeek-V4.1-Flash's performance metrics suggest superior reasoning and efficiency

But the "psyop" narrative ignores the technical rigor of open-source contributions. The real panic should be about the potential for misuse, not the model itself.

Emilio Ranucoli - RanukDEVSep 19

One app just put GPT-6, Claude, Gemini, Grok, Kimi and DeepSeek behind a single search bar with zero token limit

Someone opened their laptop and typed "search models" like they were browsing a menu, except the menu had every frontier lab on it at once. Scroll the dropdown and

bluumikSep 194

DeepSeek-v4.1-flash is really good with svg

Here's a turntable, all made one-shot. Crisp quality needs appreciation. Again one-shot, high thinking.

AJSep 1911