beating
3 stories and discussions about beating, aggregated from every source we track.
1.
How our Kimi K3 megakernel on TPU v7 reaches over 700 tokens/s with speculative decoding and nearly 2× GB200's batch-one decode throughput.
2.
An AI coding agent that doesn't read - it queries. 391 of 500 real GitHub issues resolved at 9.5 cents per fix.
3.