DedeProGames/open-reasoner

DedeProGames/ open-reasoner

57 Downloads Updated 2 weeks ago

OpenReasoner is a fine-tuned qwen3:8b and qwen3:1.7b model trained on the OpenThoughts-114k dataset.

tools thinking 1.7b 8b

ollama run DedeProGames/open-reasoner:1.7b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "DedeProGames/open-reasoner:1.7b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='DedeProGames/open-reasoner:1.7b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'DedeProGames/open-reasoner:1.7b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Applications

Claude Code

Claude Code ollama launch claude --model DedeProGames/open-reasoner:1.7b

Codex

Codex ollama launch codex --model DedeProGames/open-reasoner:1.7b

OpenCode

OpenCode ollama launch opencode --model DedeProGames/open-reasoner:1.7b

OpenClaw

OpenClaw ollama launch openclaw --model DedeProGames/open-reasoner:1.7b

Models

Name

2 models

Size

Context

Input

open-reasoner:1.7b

1.4GB · 40K context window · Text · 2 weeks ago

open-reasoner:1.7b

1.4GB

40K

Text

open-reasoner:8b

5.2GB · 40K context window · Text · 3 weeks ago

open-reasoner:8b

5.2GB

40K

Text

Readme

OpenReasoner is a reasoning-first family of compact models designed for deep, step-by-step problem solving on everyday hardware. Despite its relatively small parameter count, OpenReasoner aims to deliver strong “slow-thinking” performance with practical latency and memory usage for on-device and low-resource setups.

OpenReasoner models are fine-tuned on OpenThoughts-114k — a dataset derived by distilling DeepSeek-R1, using the public data pipeline available on GitHub (see the OpenThoughts-114k dataset card for details).

Models

open-reasoner:8b

Base: Qwen3 (8B-class)
Goal: Best overall reasoning quality in the OpenReasoner family
When to use: Math, logic, code reasoning, multi-step tasks, long-form answers
Typical hardware target: Works best with a decent GPU, but can run on CPU (slower).
(Exact speed depends on quantization, context length, and your machine.)

open-reasoner:1.7b

Base: Qwen3 (1.7B-class)
Goal: Lightweight reasoning you can run on weaker devices
When to use: Fast iteration, assistants on older PCs/laptops, smaller VRAM setups
Typical hardware target: Much easier to run on modest GPUs and even CPU-only.

Benchmarks

Model	AIME24	MATH500	GPQA-Diamond	LCBv2 Easy	LCBv2 Medium	LCBv2 Hard	LCBv2 All
OpenThinker-7B	31.3	83.0	42.4	75.3	28.6	6.5	39.9
Bespoke-Stratos-7B	22.7	79.6	38.9	71.4	25.2	0.8	35.8
DeepSeek-R1-Distill-Qwen-7B	60.0	88.2	46.9	79.7	45.1	14.6	50.1
gpt-4o-0513	8.7	75.8	46.5	87.4	42.7	8.9	50.5
o1-mini	64.0	85.6	60.0	92.8	74.7	39.8	72.8
open-reasoner:8b	76.0	97.4	62.0	—	—	—	—

open-reasoner:8b

ollama pull DedeProGames/open-reasoner:8b

open-reasoner:1.7b

ollama pull DedeProGames/open-reasoner:1.7b

![Captura de tela 2026-02-17 192811 (1).png](/assets/DedeProGames/open-reasoner/432035a5-b343-4794-b114-6e702289a2c9)
OpenReasoner is a **reasoning-first** family of compact models designed for **deep, step-by-step problem solving on everyday hardware**. Despite its relatively small parameter count, OpenReasoner aims to deliver **strong “slow-thinking” performance** with practical latency and memory usage for on-device and low-resource setups.

OpenReasoner models are fine-tuned on **[OpenThoughts-114k](https://huggingface.co/datasets/open-thoughts/open-thoughts-114k)** — a dataset derived by distilling **DeepSeek-R1**, using the public data pipeline available on GitHub (see the OpenThoughts-114k dataset card for details).

---

## Models

### **open-reasoner:8b**
- **Base:** Qwen3 (8B-class)
- **Goal:** Best overall reasoning quality in the OpenReasoner family
- **When to use:** Math, logic, code reasoning, multi-step tasks, long-form answers
- **Typical hardware target:** Works best with a decent GPU, but can run on CPU (slower).  
  *(Exact speed depends on quantization, context length, and your machine.)*

### open-reasoner:1.7b
- **Base:** Qwen3 (1.7B-class)
- **Goal:** Lightweight reasoning you can run on weaker devices
- **When to use:** Fast iteration, assistants on older PCs/laptops, smaller VRAM setups
- **Typical hardware target:** Much easier to run on modest GPUs and even CPU-only.

---

## Benchmarks

| Model | AIME24 | MATH500 | GPQA-Diamond | LCBv2 Easy | LCBv2 Medium | LCBv2 Hard | LCBv2 All |
|---|---:|---:|---:|---:|---:|---:|---:|
| OpenThinker-7B | 31.3 | 83.0 | 42.4 | 75.3 | 28.6 | 6.5 | 39.9 |
| Bespoke-Stratos-7B | 22.7 | 79.6 | 38.9 | 71.4 | 25.2 | 0.8 | 35.8 |
| DeepSeek-R1-Distill-Qwen-7B | 60.0 | 88.2 | 46.9 | 79.7 | 45.1 | 14.6 | 50.1 |
| gpt-4o-0513 | 8.7 | 75.8 | 46.5 | 87.4 | 42.7 | 8.9 | 50.5 |
| o1-mini | 64.0 | 85.6 | 60.0 | **92.8** | **74.7** | **39.8** | **72.8** |
| **open-reasoner:8b** | **76.0** | **97.4** | **62.0** | — | — | — | — |

---

open-reasoner:8b
```bash
ollama pull DedeProGames/open-reasoner:8b
```

open-reasoner:1.7b
```bash
ollama pull DedeProGames/open-reasoner:1.7b
```

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