pdurlej/ qwythos-9b-claude-mythos-5-1m:Q4_K_M

898 Downloads Updated 1 month ago

Claude-style reasoning, vision and tool use in a compact 9B model built to run locally.

vision

ollama run pdurlej/qwythos-9b-claude-mythos-5-1m:Q4_K_M

curl http://localhost:11434/api/chat \
  -d '{
    "model": "pdurlej/qwythos-9b-claude-mythos-5-1m:Q4_K_M",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='pdurlej/qwythos-9b-claude-mythos-5-1m:Q4_K_M',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'pdurlej/qwythos-9b-claude-mythos-5-1m:Q4_K_M',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 month ago

1 month ago

6efe22249827 · 6.8GB ·

model

archqwen35

·

parameters9.2B

·

quantizationQ4_K_M

5.9GB

projector

archclip

·

parameters456M

·

quantizationF16

918MB

Readme

Qwythos 9B Claude Mythos 5 — Q4_K_M

Claude-style reasoning, vision and tool use in a compact 9B model built to run locally.

Qwythos is a Qwen3.5-9B reasoning model trained on Claude Mythos/Fable reasoning traces. It supports thinking, tool calling, image input and a context window of up to 1M tokens.

Best for

Reasoning, coding and research
Document and image understanding
Tool use and agentic workflows
Long-context experimentation

Model details


Quantization	Q4_K_M
Download size	6.8 GB, including the vision projector
Context	Up to 1M tokens
Input	Text and images
Capabilities	Thinking, tools and vision

Run

ollama run pdurlej/qwythos-9b-claude-mythos-5-1m:Q4_K_M

Recommended settings

Temperature: 0.6
Top P: 0.95
Top K: 20
Repeat penalty: 1.05

Start with a 16K context window and increase it only when needed. The full 1M-token context requires substantial memory and is not practical on entry-level MacBooks.

MacBook requirements

Usage	Practical recommendation
Minimum	Apple Silicon MacBook with 16GB unified memory; text-only use at 4K–16K context
Recommended	MacBook Air or Pro with 24GB–32GB; vision, tools and 16K–64K context
Large-context experiments	MacBook Pro with 48GB–128GB; actual limits depend on context and other running apps

Allow approximately 10GB of free storage. A MacBook Pro offers active cooling and higher sustained performance than a fanless MacBook Air.

Notes

Qwythos is a reasoning model and may use a significant portion of its output budget for thinking. Verify model output before using it in safety-critical workflows.

Original model and credits: empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF

# Qwythos 9B Claude Mythos 5 — Q4_K_M

**Claude-style reasoning, vision and tool use in a compact 9B model built to run locally.**

Qwythos is a Qwen3.5-9B reasoning model trained on Claude Mythos/Fable reasoning traces. It supports thinking, tool calling, image input and a context window of up to 1M tokens.

## Best for

- Reasoning, coding and research
- Document and image understanding
- Tool use and agentic workflows
- Long-context experimentation

## Model details

| | |
|---|---|
| Quantization | Q4_K_M |
| Download size | 6.8 GB, including the vision projector |
| Context | Up to 1M tokens |
| Input | Text and images |
| Capabilities | Thinking, tools and vision |

## Run

```bash
ollama run pdurlej/qwythos-9b-claude-mythos-5-1m:Q4_K_M
```

## Recommended settings

- Temperature: `0.6`
- Top P: `0.95`
- Top K: `20`
- Repeat penalty: `1.05`

Start with a 16K context window and increase it only when needed. The full 1M-token context requires substantial memory and is not practical on entry-level MacBooks.

## MacBook requirements

| Usage | Practical recommendation |
|---|---|
| Minimum | Apple Silicon MacBook with 16GB unified memory; text-only use at 4K–16K context |
| Recommended | MacBook Air or Pro with 24GB–32GB; vision, tools and 16K–64K context |
| Large-context experiments | MacBook Pro with 48GB–128GB; actual limits depend on context and other running apps |

Allow approximately 10GB of free storage. A MacBook Pro offers active cooling and higher sustained performance than a fanless MacBook Air.

## Notes

Qwythos is a reasoning model and may use a significant portion of its output budget for thinking. Verify model output before using it in safety-critical workflows.

Original model and credits: [empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF](https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF)

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