---
license: apache-2.0
language: [sr, hr, bs, mk, sl, sq, cnr, bg, el, tr, ro, hu]
base_model: Qwen/Qwen3-8B
library_name: transformers
tags: [balkan, southeast-europe, multilingual, honest-ai, sovasoft]
---

🌐 [EN](README.md) · [SR](README_sr.md) · [HR](README_hr.md) · [BS](README_bs.md) · [MK](README_mk.md) · [SL](README_sl.md) · [SQ](README_sq.md) · [CNR](README_cnr.md) · [BG](README_bg.md) · [EL](README_el.md) · [TR](README_tr.md) · [RO](README_ro.md) · [HU](README_hu.md)


<p align="center">
  <img src="images/Zora_Stich_1700_SW_Print.png" width="520" alt="Zora — Göttin der Morgenröte, umgeben von den Symbolen der 12 Völker"/>
</p>

# Zora v1.11 — an open, honest LLM for the Balkans & Southeast Europe

*зора = „dawn". One to unite them all.* — by **Sovasoft** ([ai.in.rs](https://ai.in.rs))

---

## 1 · What Zora is
Zora is an **open 8B language model** (built on Qwen3-8B) for **12 languages of the Balkans and
Southeast Europe**: Serbian, Croatian, Bosnian, Macedonian, Slovenian, Albanian, Montenegrin,
Bulgarian, Greek, Turkish, Romanian, Hungarian.

Zora is not built to be the biggest model — it is built to be **honest, in-language, and multi-perspective**:
- **thinks in the target language** instead of pivoting through English,
- shows **several perspectives** on contested topics instead of one national view,
- and above all: **admits when it doesn't know** instead of inventing facts.

## 2 · The development story (v1.0 → v1.1 → v1.11)
| Version | Languages | State |
|---|---|---|
| **v1.0** | 6 | first public release |
| **v1.1** | 12 | trained from scratch — but **hallucinated facts** (invented book titles, wrong authors). **Never released.** |
| **v1.11** | 12 | the honest fix: says „I don't know", searches when unsure, in-language reasoning. **This release.** |

v1.1 taught us the key lesson — a small model can't *memorize* every fact, so instead of faking it,
**v1.11 was retrained to be honest** (see the benchmark below).

## 3 · Strengths & limits (honest)
**Strengths**
- ✅ **Honesty:** admits uncertainty instead of hallucinating (HALLU 1→10, DETAIL 0→8 vs v1.1)
- ✅ **In-language quality:** teaching, long-form, instructions — all strong across 12 languages
- ✅ **Tool discrimination:** answers general knowledge directly, searches only when needed
- ✅ **Multi-perspective** on contested history/culture

**Limits (be aware)**
- ⚠️ **Multi-step logic/math** (LOGIC/ANALYSIS) is weak — an 8B limit; step-by-step `<think>` helps and improves in v1.12
- ⚠️ **Factual depth is limited** (8B) — for exact facts (law, dates, current data) use **RAG/web-search**, don't rely on memory
- ⚠️ **Quantization:** use **Q5_K_M / Q6_K / Q8_0**. **Avoid Q4 and below** — heavy quantization made the model hallucinate in our tests.

## 4 · Benchmark (BalkanBench, 13 axes × 12 languages)

🔬 **BalkanBench is open** — test any model yourself: https://github.com/olivilo/balkanbench
Deterministic scoring (script / language / keywords / numbers). v1.11 (16-bit): **84/156**.

**Zora leads the field** — beating models 3–4× its size, including the current
Gemma-4-31B and Qwen3.6-30B:

![Overall ranking](charts/rank_en.png)

| Model | Size | Score |
|---|---|---|
| **Zora v1.11** | **8B** | **84** |
| Gemma-4-31B | 31B | 77 |
| Mistral-24B | 24B | 73 |
| Qwen3.6-30B | 30B | 73 |
| Salamandra | 7B | 66 |
| EuroLLM | 9B | 65 |
| Qwen2.5-32B | 32B | 65 |
| Gemma-2-27B | 27B | 63 |
| Aya | 8B | 61 |
| BgGPT | 7B | 56 |
| YugoGPT | 7B | 35 |

Quantization stays strong — Q5/Q6/Q8 all remain usable (avoid Q4):


The honesty fix in numbers, and per-axis strengths:

![Evolution v1.1 → v1.11](charts/evolution_en.png)
![v1.11 axis matrix](charts/matrix_en.png)

## 5 · Usage
**Recommended quant:** Q5_K_M (balanced) or Q6_K (near-lossless). Ollama: `ollama run olivilo/zora  # or: ollama run hf.co/sovasoft/zora-v1.11:Q5_K_M`.
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("sovasoft/zora-v1.11")
model = AutoModelForCausalLM.from_pretrained("sovasoft/zora-v1.11", device_map="auto")
```

📚 **Give Zora your own knowledge:** see [RAG-GUIDE](RAG-GUIDE.md) — connect your own documents (`zora_rag.py`, Ollama+numpy) or live web search. Zora answers only from the sources and admits when it cannot find the answer.

## 6 · Help Zora grow
Zora's weakest areas are factual depth and some smaller languages. **Send us open, licensable
sources** (texts, corpora, glossaries) in any of the 12 languages → **kontakt@ai.in.rs**. See the
multilingual manifesto *WHY-LANGUAGES* on why thinking in your own language matters.

## 7 · Acknowledgements
Zora exists because of open source. We give our formal, heartfelt thanks:

- **Above all, to the Qwen team at Alibaba** — for developing and open-sourcing **Qwen3** (Apache-2.0),
  the foundation model Zora is built upon. Without their generosity, Zora would not exist.
- To the **platforms and structures** that made this possible — **Kaggle, Google Colab, Modal,
  HuggingFace, Ollama, Unsloth** — for the compute, the tools, and the open infrastructure.
- To the **open-source community**, for the models, code, and knowledge freely shared with everyone.
- To the **people of the Balkans** — whose languages, voices, stories and perspectives are Zora's very heart.
- And to **all that is.**

*зора — the dawn belongs to everyone.*

---

*Sovasoft · ai.in.rs · one to unite them all*