Do Diffusion LLMs Scale? Testing Mercury 3 + Decide
Inception's Mercury 3 is a diffusion LLM running at 1,000+ tokens/sec, and Mercury Decide is a decision model that can now read images. Here's how they compare, what the independent evals say, and how fast they really are. Thanks to Inception for sponsoring this video. Links: Try Mercury Decide free on OpenRouter: https://openrouter.ai/inception/mercury-decide:free Mercury 3 playground: https://chat.inceptionlabs.ai/ Inception API platform + docs: http://platform.inceptionlabs.ai/ Mercury 3 launch post: https://www.inceptionlabs.ai/blog/introducing-mercury-3 Sam's 2nd Channel: https://www.youtube.com/channel/UCUF7WtR4JpVbQqcQGrLeADw Twitter: https://x.com/Sam_Witteveen 🕵️ Interested in building LLM Agents? Fill out the form below Building LLM Agents Form: https://drp.li/dIMes 👨💻Github: https://github.com/samwit/llm-tutorials ⏱️Time Stamps: 0:00 Intro 0:23 Diffusion LLMs 0:41 Mercury 3 Model 1:38 How diffusion LLMs work 4:35 Benchmarks vs GPT-6 Luna & Claude Haiku 5.5 5:06 Swarms of sub-agents 5:39 Speed: 1,068 tokens per second 6:15 OcTen: search on smart glasses 7:07 TinyFish, OpenAI compatibility & 1M context 7:48 Mercury Decide Model 8:38 Independent evals 9:39 Vision + the System 1 / System 2 pattern 9:58 Demo: watching diffusion generate text 12:29 Real-time speed test 12:52 5,000-word essay test 13:50 Agentic tool-calling tests 14:49 RAG test 16:13 Mercury Decide with images 19:28 Pipeline: Mercury Decide + Mercury 3 21:21 Launch pricing
