Notes /
Kimi K3 vs DeepSeek v4: the one-shot test
Same one-page showcase test, clearly different league.
Previously posted my one-shot web page test of DeepSeek v4:
https://v4-one-shot-demo.madeby.dev
It was okay-ish.
Yesterday I ran the exact same test with Kimi K3: a single prompt written by Kimi itself, same questions, one-page presentation.
Kimi’s output is clearly on another level. K3 is now the LLM driving my Hermes agent. Good that Moonshot lets you connect your agent (Claw or Hermes) via subscription mode, because K3 API pricing bites (only Max effort exists right now, which justifies the pricing).
K3 result:
Also posted on LinkedIn ↗