pdurlej/ gemma-4-26B-A4B-it-heretic

348 Downloads Updated 1 month ago

A fast, capable and reduced-refusal Gemma 4 MoE for creative work, coding and unrestricted local exploration.

tools thinking

ollama run pdurlej/gemma-4-26B-A4B-it-heretic

curl http://localhost:11434/api/chat \
  -d '{
    "model": "pdurlej/gemma-4-26B-A4B-it-heretic",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='pdurlej/gemma-4-26B-A4B-it-heretic',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'pdurlej/gemma-4-26B-A4B-it-heretic',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Applications

Claude Code

Claude Code ollama launch claude --model pdurlej/gemma-4-26B-A4B-it-heretic

OpenCode

OpenCode ollama launch opencode --model pdurlej/gemma-4-26B-A4B-it-heretic

Hermes Agent

Hermes Agent ollama launch hermes --model pdurlej/gemma-4-26B-A4B-it-heretic

OpenClaw

OpenClaw ollama launch openclaw --model pdurlej/gemma-4-26B-A4B-it-heretic

Models

Name

2 models

Size / Usage

Context

Input

gemma-4-26B-A4B-it-heretic:latest

17GB · 256K context window · Text · 1 month ago

gemma-4-26B-A4B-it-heretic:latest

17GB

256K

Text

gemma-4-26B-A4B-it-heretic:Q4_K_M

17GB · 256K context window · Text · 3 months ago

gemma-4-26B-A4B-it-heretic:Q4_K_M

17GB

256K

Text

Readme

Gemma 4 26B-A4B IT Heretic — Q4_K_M

A fast, capable and reduced-refusal Gemma 4 MoE for creative work, coding and unrestricted local exploration.

This reduced-refusal variant of Gemma 4 26B-A4B uses a Mixture-of-Experts architecture, activating only part of the model for each token to balance capability and generation speed.

Best for

Creative writing and roleplay
Coding and technical assistance
Reasoning and brainstorming
Tool-based local assistants

Model details


Architecture	Gemma 4 Mixture-of-Experts
Quantization	Q4_K_M
Download size	17 GB
Context	Up to 256K tokens
Input	Text
Capabilities	Thinking and tools

Run

ollama run pdurlej/gemma-4-26B-A4B-it-heretic:Q4_K_M

Start with an 8K–16K context window and increase it gradually. The full 256K context requires considerably more memory than the model weights alone.

MacBook requirements

Usage	Practical recommendation
Minimum	Apple Silicon MacBook with 24GB unified memory; close heavy apps and use 4K–8K context
Recommended	MacBook Pro with 32GB–48GB; comfortable use at approximately 16K–32K context
Larger contexts and multitasking	MacBook Pro with 64GB or more unified memory

Allow approximately 22GB of free storage. A MacBook Air can run the model, but a MacBook Pro offers active cooling and higher sustained performance.

Safety note

This model has reduced refusal behavior. Apply your own safety, privacy and content controls when using it in applications or sharing its output.

Original model: coder3101/gemma-4-26B-A4B-it-heretic

# Gemma 4 26B-A4B IT Heretic — Q4_K_M

**A fast, capable and reduced-refusal Gemma 4 MoE for creative work, coding and unrestricted local exploration.**

This reduced-refusal variant of Gemma 4 26B-A4B uses a Mixture-of-Experts architecture, activating only part of the model for each token to balance capability and generation speed.

## Best for

- Creative writing and roleplay
- Coding and technical assistance
- Reasoning and brainstorming
- Tool-based local assistants

## Model details

| | |
|---|---|
| Architecture | Gemma 4 Mixture-of-Experts |
| Quantization | Q4_K_M |
| Download size | 17 GB |
| Context | Up to 256K tokens |
| Input | Text |
| Capabilities | Thinking and tools |

## Run

```bash
ollama run pdurlej/gemma-4-26B-A4B-it-heretic:Q4_K_M
```

Start with an 8K–16K context window and increase it gradually. The full 256K context requires considerably more memory than the model weights alone.

## MacBook requirements

| Usage | Practical recommendation |
|---|---|
| Minimum | Apple Silicon MacBook with 24GB unified memory; close heavy apps and use 4K–8K context |
| Recommended | MacBook Pro with 32GB–48GB; comfortable use at approximately 16K–32K context |
| Larger contexts and multitasking | MacBook Pro with 64GB or more unified memory |

Allow approximately 22GB of free storage. A MacBook Air can run the model, but a MacBook Pro offers active cooling and higher sustained performance.

## Safety note

This model has reduced refusal behavior. Apply your own safety, privacy and content controls when using it in applications or sharing its output.

Original model: [coder3101/gemma-4-26B-A4B-it-heretic](https://huggingface.co/coder3101/gemma-4-26B-A4B-it-heretic)

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