BLAST RADIUS

Economics and capital · Quantization efficiency

Inferact Reports Faster Kimi K3 TPU Inference

Sep 23, 2026

Inferact reported 709 tokens per second for Kimi K3 on 16 TPU v7 chips versus 452 on 16 GB200 GPUs in a low-concurrency test.

Read the original at runtimewire.comOpens the publisher's site in a new tab

More in Economics and capital