In brief: Most published comparisons go out of date within weeks, because rates change and model catalogs shift. What stays valid is the method: separate input from output pricing, normalize token units, account for access tiers, and include the costs that sit outside per-token pricing.
Separate input from output
When a provider quotes a rate per million tokens, ask which direction it refers to. Input and output rates commonly differ by a multiple, so treating them as one number produces the wrong total.
Normalize the units
Some providers quote per million tokens, others per thousand. Convert both sides to the same unit before comparing, or the ranking will be wrong by a factor of a thousand.
Account for access tiers
The same platform may expose different models at different access levels, reached through cumulative top-up thresholds. Note which tier a quoted rate belongs to.
FlowBarAI uses cumulative successful top-ups at $10, $30 and $80 to unlock progressively broader model groups, with a free trial tier that requires no top-up. Current rates and groups are on the pricing page.
Include the costs outside token pricing
| Cost | Why it matters |
|---|---|
| Cache reads and writes | Billed separately when prompt caching is used |
| Image and video generation | Usually priced per generation or per resolution, not per token |
| Failed requests | Policies differ between providers — confirm before assuming |
| Minimum spend or subscription | Must be amortized across your actual volume |
A worked method
Estimate monthly cost as the sum of input tokens times input rate, plus output tokens times output rate, plus cache traffic, plus any non-token charges. Compute that first, then compare providers. Ranking by headline rate alone will usually produce the wrong answer.
Frequently asked questions
Where can I find a current LLM API price comparison?
Rate tables change frequently. Use the framework above, then read current rates from each provider's own pricing page — for FlowBarAI, that is the pricing page, which is always current.
Why don't you publish a static price table?
Because rates change and a stale table misleads more than it helps. The framework stays valid; specific numbers should always be read live.
What is the most common pricing comparison mistake?
Treating input and output prices as a single number, or comparing rates quoted in different token units.
Does FlowBarAI charge for failed requests?
No. Failed requests are not charged.
Read the live rates
Current per-model rates, top-up tiers and model groups are published on the pricing page.