pthinc/mutant_reasoning

---
license: apache-2.0
language:
- en
pretty_name: >-
MUTANT(GPT5-GROK3and4-KimiK2-GLM4.6-MinimaxM2-GPT5Codex-Gemini3Pro-Opus4.5and4.6)THINKING
task_categories:
- text-generation
- question-answering
tags:
- reasoning
- thinking
- think
- prometech
- gpt 5
- grok 3
- grok 4
- kimi k2
- glm 4.6
- minimax m2
- gpt 5 codex
- gemini 3 pro
- opus 4.6
- opus 4.5
- opus
- synthetic
- think
- mutant
- dataset
- high-level-dataset
- english
- code
size_categories:
- 1K<n<10K
---

# MUTANT REASONING DATASET

Because this dataset is qualitative rather than quantitative, it can create a 15-40% increase in the overall capability of the model (especially on the logic and coding side). It allows the model to move away from being "talkative" and become "insightful".
- When using this "mutant" dataset, you should think of it not as a mere pile of information, but as a "master class" that will elevate your model to the next level. Because the data is already very dense and high-quality, choosing a slow learning rate during training and avoiding over-repetition—that is, avoiding excessive rote memorization—is crucial. Your goal should be to instill in the model the "logical reasoning method" that these massive models follow when solving complex problems, rather than trying to cut short the answers during training. Therefore, allow those deep thought chains to flow naturally. Even just one or two rounds of training will be enough for your model to stop rambling and start acting like a sensible and wise "companion" that actually produces solutions.

## COMBINES
- reedmayhew/Grok-3-100x
- TeichAI/brainstorm-v3.1-grok-4-fast-200x
- TeichAI/kimi-k2-thinking-250x
- Liontix/minimax-m2-250x
- TeichAI/glm-4.6-250x
- TeichAI/gpt-5-codex-250x
- TeichAI/gemini-3-pro-preview-high-reasoning-1000x
- TeichAI/claude-4.5-opus-high-reasoning-250x
- TeichAI/gpt-5.1-high-reasoning-1000x
- TeichAI/Claude-Opus-4.6-Reasoning-927x

# Usage
You can easily load this dataset using the Hugging Face `datasets` library or `pandas`.

## Using Hugging Face Datasets
```python
from datasets import load_dataset

dataset = load_dataset("jsonl", data_files="mutant_reasoning.jsonl")
print(dataset['train'][0])
```

## Using Pandas
```python
import pandas as pd

df = pd.read_json("mutant_reasoning.jsonl")
print(df.head())
```

## License
This dataset is released under the **Apache 2.0** license. Please refer to the original source datasets for their specific licensing terms where applicable.

*© 2026 Prometech A.Ş. - All Rights Reserved.*