← Back to CreditFX

How it works

The methodology behind the rates, the leaderboards, and what gets excluded.

1. Every rate is normalized to a real unit

A platform's own pricing page tells you what a subscription costs and how many "credits" it includes. That's not useful on its own — a credit means something different everywhere. CreditFX does the conversion for you: it divides the monthly cost by the credits included, then multiplies by how many credits a unit of output actually takes, landing on a plain rate like $/second of video, $/minute of audio, $/image, or $/million tokens.

Runway's Standard plan: $12/month for 625 credits. Its Gen-4.5 model costs 12 credits per second of video. $12 ÷ 625 × 12 = $0.23 per second.

2. Billing models are tagged, not blended

Not every platform charges the same way, and treating them as if they did would produce misleading numbers. Every platform on the board carries one of four tags:

Credit-meteredSubscription + credit-gated AIFlat / no creditsPay-per-use API

A platform charging a flat $25/seat/month (no credits at all) is a fundamentally different product than one metering every second of output — so they're never ranked head-to-head as if one were simply "cheaper" than the other.

3. Leaderboards only rank what's genuinely comparable

A "best value" ranking is only shown when two or more platforms share the same real unit. Seconds and minutes of video get converted onto one shared basis so generative tools and avatar tools can be ranked fairly against each other — but a platform priced per seat, per message, or per completed task doesn't get forced into that same ranking. It's listed separately instead, with a note explaining why.

Some categories naturally split further. In LLM APIs, for example, a handful of companies train their own models (OpenAI, Anthropic, DeepSeek) while others just rent out fast hosting for someone else's open-weight model (Groq, Cerebras). Both get their own leaderboard rather than one misleading combined list — paying for speed on a rented model isn't the same value proposition as paying for a company's own frontier model.

4. Every tier, every creation mode — where we have the data

Platforms with multiple pricing tiers show all of them, not just one representative plan — pricing can shift meaningfully by tier, and sometimes counterintuitively (a bigger plan isn't always the better per-unit deal). Platforms with more than one way to create — like text-to-video versus a fully automated generation mode — let you pick the specific method, since the credit cost can differ 40x or more between them on the same platform.

This level of detail exists for the platforms we've researched most deeply so far. Others still show a single representative rate while we work through the rest — that's flagged directly in the calculator when it applies.

5. Pricing changes. Ours might lag.

Every rate carries the date it was last checked. AI platforms change pricing often, sometimes without warning — treat every number here as a strong starting point, not a guarantee, and confirm current pricing directly with the platform before budgeting anything significant.

See something off?

If a rate looks wrong or out of date, let us know on the Support page — corrections keep this useful for everyone.