Windows 10/11
x64 desktop app, no account needed
Official BenchQuorum website
CPU benchmark you can repeat, log and compare
BenchQuorum 2.9.26 is a Windows agent that writes your test script, runs it in the shell, records the real output in a CSV file and explains what changed between two runs.
Write bench.py, run it, add the median to results.csv and compare it with last week's row.
- write_file
bench.pysaved - Shell
python bench.pyexit 0 - append_file
results.csvrow added - read_file
results.csvread - grep
median results.csvmatches - edit_file
notes.mddiff shown
Waiting for your OK before the next stepRejectAccept
2.9.26
Released 2 Nov 2026
About 34 MB
Portable single .exe is about 25 MB
3 modes
Auto-run, Ask first, Plan only
No telemetry
Chats and API keys stay on your PC
13 languages
This site comes in 20 languages
How to run a CPU benchmark with results you can repeat
A CPU benchmark is fair only when every run uses the same program version, input and power plan. Write all three down. Do one warm-up run that you do not count, then repeat the test at least 5 times and report the median, because the single best run hides how much your results move from one run to the next.
Make a fair CPU comparison before a BIOS update
A fair comparison needs the same script before and after the update. Ask BenchQuorum to write bench.py, which makes one warm-up call, times 5 runs and prints the median. The agent uses write_file to create the script and Shell to run python bench.py, and it judges success by the real exit code. Then append_file adds the median, the date and the setting name to results.csv. After the update you repeat the run and the agent lays both rows in one table. Nothing is accepted blindly. Every file change appears as a diff, and in Ask first mode each write waits for your OK.
Tool trace
write_file bench.py → saved
Shell python bench.py → exit 0
append_file results.csv → row added
read_file results.csv → readWrite down the conditions of every run
Results are comparable only when the conditions match, so record them next to each run. Ask the agent to run systeminfo in Shell, which prints the Windows and hardware summary, and to save it with write_file as system.txt. It then drafts conditions.md as a checklist: power plan, laptop plugged in or not, background programs closed, similar room temperature. You fill the gaps. Only you can see the room. Both files sit in the same folder as the 5 runs they describe, and you read the edit_file diff before accepting any change.
Tool trace
Shell systeminfo → exit 0
write_file system.txt → saved
write_file conditions.md → saved
edit_file conditions.md → diff shownCheck a long run for throttling
A hot CPU can lower its clock speed, so a long run may score lower than a short one. Ask the agent to run bench.py twice in a row with Shell and compare the two medians in a table. While it runs, watch the CPU speed in Task Manager's Performance tab and use your monitoring tool for temperature. Heat is a clue. Tell the agent what you saw, and edit_file adds it to the notes column of results.csv. If the second median is slower, let the PC cool, repeat the 5 runs and check again.
Repeat the test every week
A weekly run shows slow drift that a single session hides. Ask the agent to run the benchmark from the command line, first by hand in the chat in Ask first mode with each step approved, and only then let Cron repeat the prompt on a schedule you set, such as every Monday. The app and the PC must be switched on then. Ask first pauses a scheduled run until you approve it. Each run appends a row to results.csv, and grep pulls the medians. Label scheduled rows. Cron cannot close your background programs or plug in your laptop, so compare at least 5 rows before you draw a conclusion.
How BenchQuorum differs from a chat assistant
A chat assistant describes a benchmark. BenchQuorum 2.9.26 runs it. It runs your script in the shell, reads the real exit code and writes the numbers it measured into a file you can open.
| Typical chat assistant | BenchQuorum | |
|---|---|---|
| Running the test | Prints a command for you to copy and run yourself | Runs it in the shell and reads the real exit code |
| Result numbers | Works from numbers you type or paste into the chat | Records only numbers measured on your PC, with the settings |
| Script changes | Shows code in the chat for you to paste | Edits the file itself and shows a diff first |
| Repeating runs | Needs you to start each run by hand | Cron repeats a saved prompt on the schedule you set |
How it works
- Install BenchQuorum on Windows 10 or 11 and connect a model: a cloud AI provider, a local model or the built-in model.
- Choose the working folder that holds bench.py, results.csv and your notes, and pick Ask first so you approve each change.
- Describe the test in plain words: which program to run, how many timed runs and where to save the median.
- Read the tool cards and diffs, check the numbers against the console output yourself, then accept or reject each change.
Who uses it and when
Home PC builders and gamers who tune their own machine use BenchQuorum to test before and after a driver or setting change. Some start with a web page that offers a PC benchmark online. This one runs locally. The agent saves the power plan and program version that you state, or that a command prints, next to each result, so old results stay clear.
Developers and admins who look after their own machines use it to check a PC before and after a change. Anyone who searches for "benchmark test pc" usually wants a routine for testing each PC the same way, with the exact commands and settings saved in a folder. Notes stay with the data.
Students and reviewers who read published charts use it to learn how a result is produced. A multi-core score says little about a program that mostly uses one core, and a single-core test shows that case. Read charts critically. The agent compares the settings you recorded for each run, so two results from 5 runs each stay comparable.
Questions and answers
What is a CPU benchmark?
A fixed, repeatable piece of work that records how long the processor takes to finish it, so two results can be compared fairly. Single-core results show how fast one task runs, and multi-core results show how much work runs in parallel. Conditions must match. Change nothing between runs except the one thing you are testing.
Is benchmarkgo.com the official BenchQuorum website?
Yes. benchmarkgo.com is the official BenchQuorum website, and the installer on the download page is the original build, so you can compare its SHA-256 checksum with the published manifest. Questions about a download can go through the Send feedback button inside the installed app.
Is a PC benchmark test online the same as a test I run on my own PC?
No. An online page runs inside your browser, while BenchQuorum runs your own test command on your PC and reads the real exit code and the numbers it prints. You still choose the program, input and power plan, and the agent writes them down with each result.
Why is the read and write speed so slow in my drive test?
The usual causes are a test file small enough to sit in cache, a nearly full drive, a laptop on battery, or a USB drive on a poor port, cable or hub, which gives a low read and write speed. Check these first. Repeat the run 5 times and compare the median before you decide.
Can BenchQuorum run a benchmark on my laptop?
Yes. BenchQuorum runs on Windows 10 and 11 laptops and desktops, and it runs your benchmark command in the shell. Plug the laptop in, keep the same power plan, close background programs and note the run length, because heat can lower the CPU clock speed in a long test.
Where do my results and prompts go?
The results stay on your PC, and BenchQuorum has no telemetry. The text of your prompts does go to the model endpoint you pick: a cloud provider sees it, while a local model keeps it on the machine. Optional features such as web search, phone control and update checks use the network. See the privacy page.
Which models can I connect?
Three kinds: a cloud AI provider through a standard API endpoint, a model running locally on your PC, or the built-in model. The Test button checks an endpoint and shows the response time and the context window. Pick one, press Test.
Run your first repeatable test
Download BenchQuorum 2.9.26, open a working folder and ask for bench.py. The setup program is about 34 MB, and a portable single .exe of about 25 MB also exists if you prefer no installer. bench.py needs Python.
Download BenchQuorum for Windows ↗Explore the site
Every page of this site is available in 20 languages.
What BenchQuorum can do on your PC ↗ Connect a cloud, local or built-in model ↗ Step-by-step benchmark workflow ↗ Download and verify the installer ↗ Answers to common questions ↗ What stays on your PC ↗
