Kimi K3 Inference Speed Lags Behind Claude Opus

Data from OpenRouter shows that Kimi K3 currently runs at about 26-28 tokens per second, only half the speed of Claude 3 Opus. Developers speculate that speculative decoding has not yet been enabled, highlighting the urgent need for inference optimization upon its official release.

2026-07-16 ~ 2026-07-17 · 3 related posts

• Kimi K3 Hits 28tks/s on OpenRouter — scaling01 · 2026-07-16
• K3 Model Hits Only 26 tps on OpenRouter — nrehiew_ · 2026-07-17
• Kimi K3 Inference Speed Revealed, Optimization Needed — nrehiew_ · 2026-07-17