Zentra

Development

0%

Skip to content
All Models
Previewcodereasoningedgemultilingual

Zentra 3B

A compact, efficient language model optimized for reasoning and code generation. Built for edge deployment and real-time applications.

3.2B params
32,768 tokens context
Transformer (Decoder-only)
2026-07-15

Technical Specifications

Parameters3.2 billion
Context32,768 tokens
ArchitectureTransformer
Layers28
Heads24
Hidden Size2,048
Vocab128,000
LicenseApache 2.0

System Requirements

Minimum
RAM8 GB
VRAM4 GB
OSLinux, macOS, Windows
Recommended
RAM16 GB
VRAM8 GB
OSLinux, macOS, Windows

About

Zentra 3B is our flagship small language model, designed from the ground up for efficiency without sacrificing capability. Trained on a diverse corpus of code, technical documentation, and reasoning tasks, it excels at:

- Code completion and generation across 30+ languages - Step-by-step mathematical reasoning - Structured data extraction and transformation - Real-time chat and assistant applications

Despite its compact size, Zentra 3B punches above its weight class, outperforming many 7B parameter models on reasoning benchmarks while running comfortably on consumer hardware.

Benchmarks
HumanEval
62.8%
MMLU
58.4%
GSM8K
54.2%
TruthfulQA
61.1%

Get Zentra

Download the GGUF weights for local inference, or try it in the playground.

Open SourceApache 2.0 LicenseHugging Face