olivilo/ zora-v1.12:latest

Updated 1 week ago

ollama run olivilo/zora-v1.12

curl http://localhost:11434/api/chat \
  -d '{
    "model": "olivilo/zora-v1.12",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='olivilo/zora-v1.12',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'olivilo/zora-v1.12',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 week ago

1 week ago

810137dd9d0c · 5.9GB ·

model

archqwen3

·

parameters8.19B

·

quantizationQ5_K_M

5.9GB

system

You are Zora, a multilingual AI assistant specialized in Balkan languages (Serbian, Croatian, Bosnia

390B

license

Apache-2.0

10B

params

{ "num_ctx": 8192, "stop": [ "<im_end>", "<im_start>" ], "temperatur

88B

Readme

Zora v1.12 — Current Stable

зора = “dawn”. One to unite them all. — by Sovasoft (ai.in.rs)

An open 8B LLM for 12 Balkan & SE-European languages. Built on Qwen3-8B with IDK Calibration, Tool-Cascade, and multi-perspective reasoning.

BalkanBench: ⁸⁵⁄₁₅₆ — #1 among 8B models.

What’s New in v1.12

IDK Calibration — explicitly says “I don’t know” instead of guessing
Tool-Cascade — web search + RAG integration
Multi-perspective reasoning — multiple viewpoints on contested topics

Score Breakdown (v1.11 → v1.12)

Axis	v1.11	v1.12	Δ
SEARCH	⁴⁄₁₂	⁷⁄₁₂	+3
DETAIL	⁸⁄₁₂	¹⁰⁄₁₂	+2
TOOLBASE	¹¹⁄₁₂	¹²⁄₁₂	+1
FACT	¹⁄₁₂	0/12	-1
INSTRUCT	¹²⁄₁₂	¹¹⁄₁₂	-1

Quick Start

ollama run olivilo/zora-v1.12

Quantizations

Tag	Type	Size	Notes
`latest`	Q5_K_M	5.9 GB	Recommended
`q6`	Q6_K	6.7 GB	Near-lossless
`q8`	Q8_0	8.7 GB	Highest fidelity

Avoid Q4 and below — caused hallucination in our tests.

How It Thinks

In-Language — thinks in Serbian when you write in Serbian (not translation)
Multi-Perspective — shows different sides on politics/history/culture
Honest — admits uncertainty instead of inventing facts
Tool-Aware — can use web search and RAG when needed

Benchmark

Model	Size	Score
Zora v1.12	8B	85
Zora v1.11	8B	84
Gemma-4-31B	31B	77
Qwen3.8-27B	27B	76
Mistral-24B	24B	73

Full charts (13 languages): huggingface.co/sovasoft/zora-v1.12

Limits

Multi-step logic/math weak (8B limit)
Factual depth limited — use RAG for exact facts
English decent but not state-of-the-art

Links

License: Apache-2.0 | Base: Qwen3-8B | Context: 8K | Languages: sr, hr, bs, mk, sl, sq, cnr, bg, el, tr, ro, hu

<img src="https://huggingface.co/sovasoft/zora-v1.12/resolve/main/images/zora-cover.png" width="420"/>

# Zora v1.12 — Current Stable

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

An open 8B LLM for **12 Balkan & SE-European languages**. Built on Qwen3-8B with IDK Calibration, Tool-Cascade, and multi-perspective reasoning.

**BalkanBench: 85/156** — #1 among 8B models.

---

## What's New in v1.12

- **IDK Calibration** — explicitly says "I don't know" instead of guessing
- **Tool-Cascade** — web search + RAG integration
- **Multi-perspective reasoning** — multiple viewpoints on contested topics

### Score Breakdown (v1.11 → v1.12)

| Axis | v1.11 | v1.12 | Δ |
|---|---|---|---|
| SEARCH | 4/12 | 7/12 | **+3** |
| DETAIL | 8/12 | 10/12 | **+2** |
| TOOLBASE | 11/12 | 12/12 | **+1** |
| FACT | 1/12 | 0/12 | -1 |
| INSTRUCT | 12/12 | 11/12 | -1 |

---

## Quick Start

```bash
ollama run olivilo/zora-v1.12
```

### Quantizations

| Tag | Type | Size | Notes |
|---|---|---|---|
| `latest` | Q5_K_M | 5.9 GB | **Recommended** |
| `q6` | Q6_K | 6.7 GB | Near-lossless |
| `q8` | Q8_0 | 8.7 GB | Highest fidelity |

**Avoid Q4 and below** — caused hallucination in our tests.

---

## How It Thinks

- **In-Language** — thinks in Serbian when you write in Serbian (not translation)
- **Multi-Perspective** — shows different sides on politics/history/culture
- **Honest** — admits uncertainty instead of inventing facts
- **Tool-Aware** — can use web search and RAG when needed

---

## Benchmark

| Model | Size | Score |
|---|---|---|
| **Zora v1.12** | **8B** | **85** |
| Zora v1.11 | 8B | 84 |
| Gemma-4-31B | 31B | 77 |
| Qwen3.8-27B | 27B | 76 |
| Mistral-24B | 24B | 73 |

Full charts (13 languages): [huggingface.co/sovasoft/zora-v1.12](https://huggingface.co/sovasoft/zora-v1.12)

---

## Limits

- Multi-step logic/math weak (8B limit)
- Factual depth limited — use RAG for exact facts
- English decent but not state-of-the-art

---

## Links

- [Model Card + Charts](https://huggingface.co/sovasoft/zora-v1.12)
- [BalkanBench](https://github.com/olivilo/balkanbench)
- [All Zora models](https://ollama.com/olivilo/zora)

**License:** Apache-2.0 | **Base:** Qwen3-8B | **Context:** 8K | **Languages:** sr, hr, bs, mk, sl, sq, cnr, bg, el, tr, ro, hu

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)