Zyphra’s ZAYA1-8B model, trained on AMD hardware, achieves competitive performance with frontier models like DeepSeek-R1 and Claude Sonnet 4.5 on math and reasoning benchmarks. Utilizing a mixture of experts architecture with only 760 million active parameters, ZAYA1-8B demonstrates efficient reasoning capabilities. While excelling in math and coding, the model shows limitations in agentic tasks and instruction following.