As most of you know, Deepseek V3 has already gone viral in the past 2 days since its launch on Boxing Day. Hunting it on ProductHunt here to share this great model for a wider audience: 💁 What is it? DeepSeek V3 is by far the best performing and best of value open source AI model, which can handle a range of text-based workloads and tasks, like coding, translating, and writing essays and emails from a descriptive prompt. 🎯 Why is it important? The model was trained for two months at a cost of US$5.58 million (2048 GPUs) with 671 billion parameters, using significantly fewer computing resources than OpenAI and Meta. It is a proof that there's still a lot of room for improvements in terms of reducing wasteful data and algorithms. Quote from Deepseek's official announcement: ⚡ 60 tokens/second (3x faster than V2!) 💪 Enhanced capabilities 🛠 API compatibility intact 🌍 Fully open-source models & papers Quote from Andrej Karpathy: "For reference, this level of capability is supposed to require clusters of closer to 16K GPUs, the ones being brought up today are more around 100K GPUs. E.g. Llama 3 405B used 30.8M GPU-hours, while DeepSeek-V3 looks to be a stronger model at only 2.8M GPU-hours (~11X less compute)." Who are they? DeepSeek was spun off in July 2023 by High-Flyer Quant, which uses AI to operate one of the largest quantitative hedge funds in mainland China. WIth the belief that "AI that will benefit all of humanity", Deepseek started to develop their 1st AI cluster in 2019, way earlier than most players in the market today. Github: https://github.com/deepseek-ai/D... X (Twitter): https://twitter.com/deepseek_ai?... Some interesting comments/reviews on X: Andrej Karpathy: https://x.com/karpathy/status/18... Deedy: https://x.com/deedydas/status/18... Sarah Guo: https://x.com/saranormous/status... Brecky Yunits: https://x.com/breckyunits/status... Kevin Xu: https://x.com/kevinsxu/status/18...

Don’t know Deep Seek? Check it out their X’s announcement! https://x.com/deepseek_ai/status...