The best linear models
—without the cost
Leman.zero delivers transformer-quality results with linear complexity. Memory efficient, blazing fast, and easy to integrate.
Start BuildingOutperforming the competition
with linear efficiency.
Leman.zero achieves state-of-the-art results among linear models while using a fraction of the memory and compute of traditional transformers.
#1
Best accuracy among linear models
1B
Tokens to train 14B model
Only 3,000 training steps
~5%
Within Qwen-14B accuracy
With far less memory usage
14B Scale
Best among linear models
Arch C benchmark at 14B scale. Leman.zero leads all linear models and closes the gap with full transformers.
Arch C benchmark (14B parameter scale). Score % — higher is better.
Long Context
Perfect recall at any length
Needle-in-a-Haystack accuracy across context lengths. Leman.zero stays near-perfect where others collapse.
NIAH-1 accuracy at 500M parameter scale. Higher is better.
Benchmarks
Leman.zero vs. the competition
1B model performance on standard benchmarks. Higher is better.
Results from 1B parameter models. See our documentation for full benchmark details.
Why Leman.zero
The best of both worlds
Transformer-quality results with linear efficiency. No compromises.
Linear Complexity
O(n) attention mechanism scales efficiently with sequence length, unlike quadratic transformers.
Memory Efficient
Process longer contexts with less GPU memory. Run larger models on smaller hardware.
Long Context Support
Generalizes beyond training length for very long sequences. No context window limitations.
OpenAI-Compatible API
Drop-in replacement for existing integrations. Switch with a single line change.
Efficient Training
Train competitive models with far fewer tokens and steps. Reduce your compute costs dramatically.
State-of-the-Art Quality
Best-in-class results among linear models on SWDE, FDA, and SQUAD benchmarks.
import OpenAI from 'openai';
// Try it for free - no API key needed
const client = new OpenAI({
baseURL: 'https://carloshurtadocomin--lemanlabs-openai-api-fastapi-app.modal.run/v1',
});
const response = await client.chat.completions.create({
model: 'leman-zero',
messages: [
{ role: 'user', content: 'Hello!' }
],
});Simple Integration
One line of code to switch
Our API is fully compatible with OpenAI's SDK. Just change the base URL and you're ready. No rewrites, no migrations, no headaches.
Ready to build with
Leman.zero?
Get started in minutes. Experience state-of-the-art linear models with an API you already know.
Start Building