Your hardware
Choose your Mac
We use your chip and memory to estimate each model’s speed and memory fit. Your choice stays in this browser.
How accurate is this speed estimate?
It is a broad estimate for planning. A complete speed test on your Mac is more accurate. A model is left out when the catalog says it needs more memory than your Mac has.
Choose a guide
What do you want to find out?
Choose one path. Your Mac choice applies to both.
New estimates are paused while the network changes quickly.
Rankings are paused because this page has not received new data for more than 3 minutes.
The last valid data remains visible. Wait for a new update before you use the ranking.
Model ranking
Estimated provider payout per hour kept ready
Choose how many models you want above. We will compare choices that fit your Mac.
Building your ranking…
We are combining your Mac speed, recent model activity, and Darkbloom’s listed prices.
This ranking updates when you change your Mac or history period.
Your model choice
Choose the models you want to estimate
Select one or more models. We check the estimated memory fit before we calculate a result.
Estimated provider payout for each hour kept ready
Each card uses a different amount of recent activity. The unit stays hourly. The 24-hour illustration multiplies that hourly range by 24.
Choose models to estimate each model slot during the same hour.
Selected model
Loading this model…
Please wait while we load the latest activity.
See how this estimate was made
Estimate details
Select a model
Select a chart row to see what was used in its estimate.
Use my own measured speed
Optional live view
What this Mac is doing now
See Darkbloom activity and recently completed work. These values update each second.
Connecting
Waiting for the first local sample…
Completed output over time
Request-finish token bursts, up to the last 60 state updates. This is not a live decode trace.
Current activity
Connecting
Process load identifies heavy work even when no request has finished.
GPU memory
—
Models ready
Waiting…
True live decode and prefill tok/s are not exposed.
Darkbloom adds the available token counter when a coordinator request finishes. The dashboard does not convert those bursts into a false live rate.
How the estimate is calculated
For one model, we use Darkbloom’s capacity data to measure the share of loaded provider time that was running. We combine that share with Darkbloom’s listed price and the estimated speed of your Mac.
For a selected model set, we estimate each model slot separately and add its contribution during the same hour. The separate public provider list describes whole-provider activity and recent model focus, but it does not set the estimated payout.
A single model is a clear leader only when its low estimate is above the high estimate for every other eligible model.
How single-model work time is measured
time the model was running ÷ time the model was loaded
Invalid updates and long gaps are left out. Missing time is not treated as idle time.
How the work value is estimated
The estimated range uses input price, output price, the model’s recent OpenRouter token pattern, Mac speed, and recent running time. The result is for one hour that the model stays ready.
Token patterns used in the estimate
OpenRouter token totals give an input and output pattern for each mapped model. A weaker mapping makes data confidence lower.
Limits of a model-set result
The public provider list reports one recent model. It does not measure how several loaded models divide a Mac’s speed. Until a matching combined-speed test is available, the estimate adds the model-slot values without a slowdown. The ranking is useful for direction, but your actual payout can differ.
Data and privacy
Network tokens in the last 24 hours
Not available Extra network context. It does not change the model result.How a clear leader is chosen
Estimated ranges must not overlap The low estimate must be above every other high estimate.Your saved information
This browser only Your Mac choice and measured speeds are not sent to the data collector.