China’s Moonshot releases a new open source model Kimi K2.5 and a coding agent

Honestly? "15 trillion tokens" is marketing noise. What matters is whether K2.5 actually works—can it code, reason, go long-context? Moonshot's play is smart (open-source for China market dominance), but training data size doesn't predict quality. Look at GPT-4—it didn't win on parameter count.

We'll know in 30 days when people run it. If the model's legit, it forces price pressure on open-source tiers everywhere.