SculptAI Engineering

What we measured

Notes from building a deterministic civilization engine with a language model inside it. Every post here carries numbers we took ourselves and the command that reproduces them.

8 August 2026 · 7 min
Kimi K3 spends your token budget thinking before it answers
A 200 OK, an empty response, and a full invoice. Measured token, cost and latency figures for Kimi K3 across all four reasoning-effort settings.
Kimi K3 · OpenRouter · LLM engineering

Kimi K3 spends your token budget thinking before it answers