Just got burned by a vendor's 'effective price' marketing. Stick to raw token math.

Hey everyone. Just had a rough lesson as a beginner. Was comparing two LLM providers for a simple chat feature. One had a lower "effective price per 1M tokens" due to some context caching discount. Sounded great!

But my actual usage pattern didn't hit the cache much. My bill was way higher than the simple 'tokens in + tokens out' math from the other vendor. Felt misleading.

Can we just stick to comparing raw input/output token prices here? For those of us just starting out, the marketing math is confusing. How do you all cut through it? I'm looking at OpenAI, Anthropic, and maybe some open-source hosted options. Main concerns are cost for simple tasks and reliability.