MicroLLM Lab lets you run seven tiny language models in a browser and measure exactly how fast and accurate they are. The tool is simple: pick a model, press a button, and watch the numbers roll. It is the kind of thing that makes you want to open your laptop and try it right now.
The lab is not a review of models. It is a sandbox. Each model gets tested against a set of objective checks, and the results are measured in milliseconds, tokens per second, and accuracy rates. The page even lets you generate a certificate tied to your device’s hardware, so you can share your score on social media.
The Models on Offer
The seven models available are small by design. They fit in a browser tab. The lab treats them with a certain honesty: a 135M model is allowed to fail. That is the measurement.
Each model runs through the same tests. The checks are objective, meaning they look for specific patterns in the output rather than judging the writing quality. The page says as much: “Objective checks (regex / exact tokens), not writing quality.” That distinction matters. A model can produce coherent text and still miss a regex check.
How the Tests Work
The tests run on the decoded text from each model. The checks happen after the model generates its response, not during the generation process itself.
The metrics are threefold:
- Runtime per test, measured in milliseconds.
- Accuracy per test, measured against the objective checks.
- Speed, measured in tokens per second, both sustained decode and peak performance.
The page keeps these numbers on your machine. The tests also use your last token count if the system has one, which helps smooth out the measurements over repeated runs.
The Certificate Feature
One of the lab’s standout features is the certificate generator. It creates a verifiable performance certificate based on your device hardware, your peak tokens per second, and your sustained decode rate. You can then share that score on social media.
This is a clever move. It turns the benchmark into a sharable result. Instead of just looking at numbers on screen, you can post a certificate that proves you ran the test on your own machine. It adds a layer of trust to the whole experience.
What the Lab Measures
The lab measures speed and accuracy against objective checks. That is the core of what it offers. It is not a subjective judgment of how a model sounds. It is a measurement of how quickly a model produces output and how often that output matches the expected pattern.
The page estimates speed using your last token count if it has one. That suggests the system keeps some state across runs, which is handy for comparing performance over time.
Why This Matters
MicroLLM Lab matters because it democratizes benchmarking. A tool like this does not require a PhD in machine learning to use. Anyone with a browser can load it, pick a model, and see how it performs. That accessibility is valuable.
It also shows how far browser-based AI has come. The lab is a small window into that shift.
What We Learn From It
The lab teaches us a few things. First, that objective measurement is possible even for complex systems. Second, that sharing benchmarks is easier than ever.
The page’s own description captures the spirit of the project: “Try 7 tiny LLM’s in the browser.” That is the whole pitch. It is a playground for curious readers who want to see how these models work in practice.
“Try 7 tiny LLM’s in the browser.”
Our View on the Tool
We think this is a worthwhile experiment. It is a free, accessible way to learn about how these models perform. The certificate feature adds a fun layer of social proof. You can show off your benchmark results to friends or post them online.
There is no downside to trying it. The models are small enough to run quickly, and the tests are objective enough to be trusted. It is a great use of ten minutes if you have them.
The Bottom Line
MicroLLM Lab is a useful, free tool for anyone curious about how these models work. It measures speed and accuracy objectively, lets you generate a certificate, and shares your results. It is a small window into a larger shift in how we interact with AI.
The lab is a reminder that the technology we use every day is getting smaller, faster, and more capable. It is a quiet demonstration of progress. And it is a lot of fun to play with.
Give it a try. You will likely learn something new about how these models behave, and you might even end up sharing your certificate on social media.
Source material: “MicroLLM Lab – Try 7 tiny LLM's in the browser,” stateofutopia.com.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.