DeepSeek V4.1 Flash 1017 is really golden.

tl;dr: love LLMs to learn things, but only closed-source models were truly reliable for me, until 1017 came up: it's great and fast for everything This is just an appreciation post made by a human seriously impressed by this iteration. Quick 3615-mylife, I really hate the sloppification of everything. But having trouble since forever to read books, LLMs helped me learn so much about subjects I gave up on. For a long time, I found that open source models weren't usable for anything. Hallucinations, dumb, it was funny but couldn't rely on them for anything. Closed source models were the only way. Then GLM 5.X appeared, and I was severely impressed. For the first time I realized that, by giving the model access to web_search tools... the model was really helpful. For a really big bill, I could have it at home. But the model was still quite slow, and Z.AI coding plan was quite fishy. Slow, and sometimes became dumb for no reason. Still used it a lot with Hermes. First deepseek flash 4.1 came out, and its speed and low cost was really impressive. But sometimes too confident when wrong, and strugling on harder tasks. Not really reliable for me. ... but the 1017 iteration holy shit... The model is great and reliable enough for everything. And god it's fast. You can brainstorm really quickly on the max effort with it. And so cheap. I think I finally reached this point where "ok I'm satisfied with this, I don't need more" you see? I only use Hermes desktop with a lot of chats, I'm quite vanilla. Sometimes I'll use hermes via TUI for coding tasks. I can give it screenshots, ask it questions, and with the Kagi MCP for web search, it never did hallucinate for me (for now, fingers crossed) I really don't feel the need for any other closed source models now. I am happy. Thanks a million to the Deepseek team and the Hermes team, I never learnt so much.