New AI Software Makes Big Models Run 3x Faster on Your PC
You can now run powerful AI at home without expensive cloud fees.
A Reddit user recently shared their experience with a new AI tool called Strata that makes running large language models (AI that can chat and answer questions) on personal computers much faster. Before Strata, they could only get 21 tokens per second (a measure of speed) from a big model called Qwen3.8-Flash-Next. After installing Strata, the same model ran at 60 tokens per second—nearly three times faster. That's the difference between a frustrating wait and a smooth conversation.
This matters because running AI on your own machine, instead of paying for cloud services, can save money and keep your data private. But there's a big trade-off: the model uses so much of the computer's memory (RAM) that there's almost nothing left for other tasks. The user couldn't run their image and video AI tools at the same time. They joked about needing to upgrade to 128GB of RAM but said their wife's requests for jewelry and trips made that impossible.
So while Strata is a huge leap for local AI, it also shows how demanding these models are. For now, you might have to choose between running one powerful AI or several smaller ones. But as hardware improves and software gets more efficient, everyday computers will likely handle more. For anyone curious about running AI at home, this is a promising step—just be ready to manage your memory carefully.
- Strata makes large AI models run about three times faster on a regular computer.
- One user got 60 tokens per second, which is fast enough for daily use.
- The catch: it uses almost all your computer's memory, so you can't run other heavy programs at the same time.
Why It Matters
Faster local AI means you can save money, keep data private, and avoid cloud fees—if your computer has enough memory.