ByteInfer Labs

Inference has a new language_

The infrastructure for fast ML inference — built from first principles.

Blaze SDK

Write scalar loops. Get optimized kernels for x86, ARM, CUDA, and Metal.

Learn more →
127.6 TFLOPS BF16 Matmul · RTX 4090
1688 tok/s Qwen 3.5 9B prefill · RTX 4090
98.7 tok/s Qwen 3.5 9B decode · Q4K GGUF
1300× Same source, different schedule

You write the math. We handle the speed.