Kimi K3 matches Opus — so why is OpenAI calling it 'very token hungry'?

moonshot ai released kimi k3, a model that benchmarks show matching anthropic's opus 4.8, attributed by experts to hardware scarcity driving architectural innovation rather than data distillation. google deepmind and openai strategists have publicly acknowledged the model's strength, though it remains token-hungry and lags behind the absolute top-tier western models like gpt-5.6 sol. this development challenges the fundamental western investment thesis that compute alone dictates capability and raises fresh skepticism regarding the effectiveness of us export controls on limiting chinese ai progress.