Oct 28, 2026
How Long Does It Take to Port a New AI Model? Minutes Now, and Falling
In 2023 we bragged that porting Llama 2 to our GPNPU processor core took only 13 engineer weeks. On the day Liquid AI released LFM3, an agent had it running on an FPGA prototype of our core in about two hours, at 94% of its decode roofline, and any licensee with Claude or GPT can do the same with our skills served over MCP. The thing that made porting slow was a human writing kernels. That stopped being the bottleneck, and it changes what matters in an AI processor.
