🔥 How fast does my MacBook Pro M5 Max run Qwen3.8-27B? 🆕 There's a new benchmarking tool and leaderboard for that. 🎯 Liquid AI + Artificial Analysis released Pipette, an open-source benchmark built for AI running on local hardware. 👇 Pipette benchmarks the inference stack, including ... 🧠 Model 🗜️ Quantization ⚙️ Runtime 💻 Device 📚 Context length So instead of wondering how fast Qwen3.8 is, now you can find out how fast Qwen Q4 is in llama.cpp on my MacBook Pro M5 Max (or other local AI machines). According to AA and Liquid, the dataset currently has ... 📊 10,000+ benchmark results 🧪 1,000+ tested configurations ✅ model × quantization × runtime × device × context 🤖 ~35 model classes 🗜️ 7 quantization levels ⚙️ llama.cpp 📱 iPhone + Android 💻 Mac + Windows PCs 🧠 Strix Halo results coming soon And it measures what matters to Local AI users... ⚡ Decode tps 🚀 Prefill speed ⏱️ Latency 💾 Peak memory 🧠 Model quality ⚠️ Caveat, current benchmark only covers ... ✅ MacBook Pro — M5 Max ✅ iPhone 17 Pro ✅ Samsung Galaxy S26 Ultra 🔗 pipette dot liquid dot ai