llmspeed
Pick hardware and a model, and feel the real speed of a local LLM before you buy
Why it made the rally
Tokens per second and time to first token are simulated from community benchmarks for 30 setups, from 8 Mac configurations and NVIDIA, AMD and Intel GPUs to a DGX Spark and an H200, across 15 models like Gemma 4, GPT-OSS and DeepSeek V4. Free and MIT licensed, made by Albin Ostman.
More like this
Prompts to production UI, using your own design system
AI · hunted by Denis
Run several Claude Code agents side by side, each in its own workspace
AI · hunted by Denis
An infinite canvas that chains AI image and video models
AI · hunted by Denis
Sitemaps and wireframes to production Next.js and Tailwind
AI · hunted by Denis
Production-ready UI from prompts, exportable to Figma, code or your IDE
AI · hunted by Denis
Discussion
Sign into join the discussion.
Quiet so far. Say what you noticed and start the thread.