125B Model Runs on a Gaming PC

🔥 What happened Strata just squeezed Qwen3.8-Flash-Next — a 125-billion-parameter model — onto a normal gaming PC. An RTX 5070 with 12 GB VRAM and 64 GB RAM is enough: 53 tokens/s at 128K context, free and open source. 💡 Why it matters Until now, a model this size needed a server. Now a $500 GPU does it: 60–95 tokens/s, faster than you can read. With 24 GB VRAM (RTX 3090), expect 100–140 tokens/s. ⚡ Our take This is the moment local AI stops being a hobbyist toy and becomes a real alternative. Anyone still paying per API call in 2025 is burning money.