Kimi K3, and what we can still learn from the pelican benchmark

This article discusses the release of Kimi K3 by Chinese AI lab Moonshot AI, a 2.8 trillion parameter model touted as the first 'open 3T-class model'. It covers self-reported benchmarks where K3 outperforms Claude Opus 4.8 and GPT-5.5 but trails Claude Fable 5 and GPT-5.6 Sol. The article details pricing ($3/$15 per million tokens), cost per task, and token efficiency. The author also uses a humorous 'pelican riding a bicycle' SVG test to evaluate the model's output quality and cost, noting that such benchmarks have become less reliable over time. The content is focused on AI model comparisons, tech industry trends, and practical testing.