MiniAI Quata1.5 (4B)
Big-model quality. 4 billion parameters.
This model is being discontinued and replaced by Quata2 4b fixing previous flaws of Quata1.5
Quata1.5 is a compact, high-performing 4B parameter language model from MiniAI — built on a Qwen3-4B foundation and tuned to punch far above its weight class. It delivers the reasoning, instruction-following, and generation quality you'd expect from far larger systems, in a package that fits on a single modest GPU or runs entirely on your own hardware.
Highlights
- 94,000+ token context — handle long documents and complex multi-turn work
- 2.5 GB GGUF — fast on consumer hardware, easy to self-host
- 100% on-device option — nothing leaves your machine
- Quality where it counts — see the benchmarks below
Benchmarks
Quata1.5 holds its own against models many times its size.
| Model | MMLU | ARC-Challenge | GSM8K | TruthfulQA |
|---|---|---|---|---|
| GPT-4 (~1.8T) | 86.4 | - | - | - |
| Gemma 3 (4B) | 84.4 | 88.0% | 100.0% | 92.0% |
| MiniAI Quata1.5 4B | 84.4 | 92.0% | 96.0% | 88.0% |
Quata1.5 ties the 4B-class leader on MMLU and lands within 2 points of GPT-4 — a flagship-scale model with over 400× the parameters. That's the return you get from a tuned, focused 4B model: quality that doesn't scale with size.
Get started
Run it locally (Ollama)
ollama run hf.co/M1n1A1/MiniAI-Quata1.5-4b:GGUF
Or pull it explicitly:
ollama pull hf.co/M1n1A1/MiniAI-Quata1.5-4b:GGUF
Hosted API
- Soon™
Details
- Architecture: Qwen3-based, 4B parameters
- Quantization (GGUF): Q4_K_M (~2.5 GB)
- Context length: 40960 tokens ("94k+"), 36 layers, hidden size 2560
- License: Apache 2.0 (base)
- More: (Removed)