Positioning
One dot per plan and model. Default axes: tokens per money against AI score.
Use the arrow keys to move between dots, Enter opens its numbers, Escape clears the selection. The same values are in the table below.
Best value right now
One row per plan, strongest model, ranked by tokens per unit paid. Only plans that do not train on your data and come with API access.
| Plan | Model | AI score | Tokens / $ | Tokens / mo | Plan price / mo |
|---|---|---|---|---|---|
All plans and models
Rates are tokens or requests per 1 paid unit, in your currency.
| Plan | Model | AI score | Tokens / $ | Requests / $ | Tokens / mo | Requests / mo | Included volume | Plan price / mo | Privacy |
|---|---|---|---|---|---|---|---|---|---|
Same model, different plans
Request rate per $1 paid for one model family across plans. Plain numbers, no podium.
Budget calculator
One subscription per account, no stacking. Enter what you pay per month and see the most tokens each single plan buys you.
Ranked by monthly tokens.
Changelog
What changed for you, newest first.
Why “60 for 10” is not the answer
The sticker value hides the real economics. Here is how plans are made comparable, every step reproducible.
Real token prices, not marketing
Every model behind a plan has published API token prices: input, output, cached-read and cached-write. That is the ground truth behind any credit or dollar-usage number.
Provider credit formulas
Providers define their own credits (GLM: (in×6.9 + cached×1.7 + out×24)/10000). Those formulas come from the official docs, never invented.
Cache-aware workload model
Coding agents hit the cache constantly. Input is priced 5% fresh + 95% cached-write, cached-read and output at their real rates, per model and workload pattern.
Show all steps
Windows are caps, not volumes
A 5-hour window is a throughput limit, not a monthly quota. Weekly credits scale to a month (×4.33); 5h caps are never multiplied into fake monthly numbers.
Fair pattern unification
Some feeds reuse one generic workload pattern. For shared model families we use the most precise per-model pattern for both plans, so a cheap pattern cannot rig the comparison.
Undisclosed stays undisclosed
If a provider hides its numbers, we say so. No invented credits, no back-calculated quotas. Honesty is a feature.
Frequently asked questions
Legal
Privacy, imprint and liability information.
Privacy
No cookies, no trackers. This site sets no cookies and makes no connections to third parties. All data is served from this GitHub Pages site itself.
Local storage (your browser only). Your preferences (language, theme, currency, visible columns) are stored in your browser's localStorage so the page can remember them between visits. This data never leaves your device. localStorage is not a cookie and is not used for tracking.
GitHub Pages hosting. This site is hosted by GitHub Pages. As with any web host, GitHub's servers process technical access data (IP address, user agent, requested files) in server logs. Please refer to GitHub's privacy policy for details on their data processing.
No analytics, no advertising. We do not use analytics tools, advertising networks, or third-party embeds.
Contact. For any privacy request, please use the contact details in the imprint.
Imprint / Provider identification
This website is operated by an individual maintainer on a non-commercial, informational basis.
Operator: Kolja Knodel
Contact: GitHub: harrytyp
As a private, non-commercial website, a full postal address is not required. If this service becomes commercial or targets the DACH region, the imprint must be extended (e.g. §5 DDG in Germany) with your full name and address.
Disclaimer
All prices, quotas, and conditions shown here are collected from public provider sources on a best-effort basis. They may change at any time. Always confirm current terms on the official provider website before purchasing.
This site is independent and not affiliated with, endorsed by, or sponsored by any provider shown. We do not sell plans and have no affiliate links.
Information is provided "as is" without warranty of any kind. We are not liable for any decisions made based on this data.