Pricing

Speed is free. The economics are Pro.

One plan for everything we measure about what a model costs to run: per token, per watt, and at every load level.

Free

For checking how fast a model is before going further.

$0

  • Full leaderboard: speed, latency, peak throughput
  • Every model page with all engine runs
  • Analysis articles and the published methodology
  • Live benchmark pipeline
Open the leaderboard

Most popular

ModelPerf Pro

Save 20%

For deciding which model, engine and GPU to run.

Monthly plan price $49$39/ month

$588 $468 billed once a year. You save $120.

  • Cost per million tokens for every model, engine and GPU
  • Energy per million tokens, tokens per watt and GPU power draw
  • Full concurrency curves for all three workload scenarios
  • Unlimited ROI and break-even calculations with PDF export
  • CSV export of the full leaderboard, paid metrics included
  • New models and re-runs included as they are published
Get Pro · $468 / year

Payment confirmed with the provider before access is granted

Custom benchmark

For a model, hardware or endpoint we have not published.

Quoted

  • Your model on the hardware you choose
  • The same suite: three workloads, full concurrency sweep
  • Performance, cost and energy report
  • Private results, or published on the leaderboard
Request a benchmark

Only need one model? Unlock it for $10 from its page on the leaderboard. Pro pays for itself from the fifth model in a month.

Compare

What you get on each plan

FeatureFreePro
Tokens per second per user
Time to first token
Peak throughput and best concurrency
Cold start time
Cost per million output tokens
Energy per million tokens and tokens per watt
GPU power at peak and idle
Concurrency curves, three workloads
Sort the leaderboard by cost and energy
ROI and break-even calculations with PDF
CSV export

Questions

Pricing questions

What does ModelPerf Pro include?
Every paid metric for every published model: cost per million tokens, energy per million tokens, tokens per watt, GPU power draw and the full concurrency curves for all three workloads. It also includes unlimited ROI calculations with PDF export and CSV export of the leaderboard.
What stays free?
Single-user speed, time to first token, peak throughput, cold start and the methodology are free for every model, with no account needed.
Does Pro renew automatically?
The checkout tells you before you pay. A prepaid plan covers one month or one year and does not renew by itself; you can extend it at any time and the new period is added after the current one. Where a renewing card subscription is offered, it renews until you cancel from your account, and you keep access to the end of the paid period.
Can I pay for just one model?
Yes. A single model can be unlocked for $10, once, including its future runs. ROI calculations are also sold in packs of 5 for $20.
How is payment confirmed?
After you pay, our server confirms the payment with the payment provider and then switches Pro on. It usually takes a few seconds. If it does not, your account page has a button to check again.
Do you benchmark private models or our own hardware?
Yes, as a custom benchmark. We run the same suite on your model, your hardware or your hosted endpoint and deliver a full report. Use the request form on the services page.

Pick the model on cost, not on a guess

Start with the free leaderboard, upgrade when you need the economics.