Model Spotlight: eeny 2.0, a 999K-Parameter Model in 1.76MB
A close look at eeny 2.0 — the smallest member of the TinyLM family, under a million parameters and small enough to ship like an image.
Meet eeny 2.0
eeny is the smallest model in the TinyLM family. The 2.0 version has 999,000 parameters and weighs about 1.76MB on disk. To put that in perspective, it is smaller than many photos on a typical web page. Yet it runs as a real language model in your browser.
Why Build Something So Small
The instinct in AI is always bigger. eeny exists to test the opposite: how much useful behavior can you pack into under a million parameters? The answer turns out to be "more than you would expect" for narrow tasks — and that makes eeny ideal for places where every kilobyte and millisecond counts.
What eeny Is For
eeny is a specialist. It is well suited to tight, repetitive jobs: short autocomplete, simple intent or sentiment classification, text tidying, and quick suggestions. Because it is so small, it loads almost instantly and responds with no perceptible lag, even on older phones.
What eeny Is Not For
We will say it plainly: eeny is not a chatbot you can ask anything. With 999K parameters it has very limited world knowledge and a narrow range. Ask it open-ended questions and it will disappoint. eeny earns its keep inside a focused application where you have shaped or fine-tuned it for one job.
Under the Hood
eeny 2.0 runs on SPRAPP's Rust-to-WASM engine with int8 SIMD math. Its tiny footprint means the whole model fits easily in browser memory and caches into IndexedDB after the first load. Subsequent visits start cold-free and work fully offline.
See It Load
The best demo of eeny's size is watching it appear. At https://ai.sprapp.com the model downloads in the time it takes to load a normal web page, then runs locally. Reload with the network off and it is already there, cached and ready.
Fine-Tuning eeny
Because eeny is so small, fine-tuning it is fast and cheap. You can adapt it with LoRA on a modest dataset to specialize it for your domain — a product catalog, a support vocabulary, a particular writing style. A specialized eeny often beats a generic larger model on its narrow task while staying tiny.
Licensing and Use
eeny is available for commercial use under TinyLM licenses ranging from $19 to $99, depending on the model and terms. For a sub-megabyte model that runs offline with no serving cost, that is an easy budget line for an app that needs private, instant text intelligence.
The Takeaway
eeny 2.0 is a proof and a tool. As a proof, it shows that under a million parameters can still be useful. As a tool, it is the model to reach for when you need something genuinely tiny that still does real work. Start small — sometimes eeny is all you need.