In addition to people pointing out your LLM got the pricing wrong,
> The pricing and "including thinking tokens" framing position it as a reasoning-capable flash model rather than just a pure speed optimization
Every Gemini model starting with 2.5 has been a reasoning model.