Chinese AI Finally Surpasses Claude Opus 4.8: Kimi K3 Released #119

It is among the world's best-performing open-weight models.

"Kimi K3" follows an "open-weight" format, where the model weights are made public for anyone to download and use. It has achieved scores in third-party evaluations that surpass the renowned Claude Opus 4.8. Engineers are particularly impressed by its frontend coding capabilities, where it has ranked first in the world against formidable competitors, proving its genuine strength.

The harsh development environment is astonishing.

This achievement was accomplished amidst ongoing U.S. GPU export restrictions. The host, YOSHIO, cannot hide his surprise that such a high-performance model has emerged despite these strict limitations. Beyond its technical performance, the API usage fees are set lower than those of existing top-tier models, which is likely to accelerate its future adoption.

There is also a cautious perspective that warrants nuance.

However, the developer, Moonshot AI, has released a level-headed comment stating that there are still areas where it has not yet reached overall parity. Furthermore, some experts have raised questions regarding the evaluation methodology. As the full disclosure of the weights approaches, a certain fact has surfaced; be sure to check the main episode for the full details. We dive deep into the future prospects brought about by this latest AI.