This is one of the main reasons to use open-weight models. You can't possibly ask Claude any security-related question without triggering alarms, and let me tell you, the bar to trigger it is VERY LOW. I have even been flagged at work for asking Claude to tell me about the PRs it opened the day before, on work that it did itself... it's that sensitive. GLM or DeepSeek running locally on the other hand will never ever give you that bs, especially the abliterated weights! Local AI will win!

@Alibaba_Qwen Qwen3.8-Flash-Next on two @NVIDIAAI DGX Sparks just got better 🚀 It inherited some of the improvements i did on the single-spark recipe and a little extra more! Default recipe 👇 ・Multi-turn fix: follow-up replies 5x faster (3.25s → 0.63s), after tool calls 2.55s → 1.98s ・Code up to +27% faster (333 → 422 tok/s at 8 streams) ・~56 tok/s prose, ~76 code (1 stream), up to ~226 prose / ~422 code (8 streams) ・Optional bit-exact decoding across both nodes ・Cached-token reporting in the API ・Safer memory default: no more running both nodes at under 1 GiB free New opt-in lane on vLLM 0.30 (./start-v030.sh) 🌟 ・Follow-ups in 0.28-0.40s, and repeated prompts hit the cache (0.35s vs 2.7s) ・Same decode speed, steadier: fixed a vLLM 0.30 default that made speed swing ±15% between runs ・One small patch instead of seven ・Optional FP8 KV: 2.6M tokens of context, 3/3 needles at 200k Tested for stability for a few hours! I will keep testing this over the next days and probably make the vLLM 0.30 script the default one in the next version of the recipe ✌️

After the new additions to the Qwen3.8-Flash-Next for a single spark, coming with similar updates for the dual recipe! Testing rn to ensure it's stable. Dropping possibly tomorrow!

This. I also cancelled my Claude subscription a few months ago. It's not only about how cheap it or not is it to run it but the fact you have total freedom: • no lobotomized models • ability to run abliterated/uncensored models • PRIVACY!! no data training on your prompts. Besides, it's cool to own hardware. Internet is gone? You still have your own intelligence!

Local AI will win. Not only because it's sovereign and private, but because not everybody needs frontier AI models on a daily basis. We gave internet to people and most just watch funny cats videos 😅 With AI, most people want to do mild research on regular day-to-day topics, generate images/videos, ask legal or financial questions, "google" stuff... You can already do that very well with a single @NVIDIAAI DGX Spark, current open-weight models are very capable!