Gemini 3 Flash Preview Reasoning Prompts for Model Distillation

A reasoning dataset containing 1.01 million tokens generated using the Gemini 3 Flash Preview model with high reasoning depth. Created by TeichAI via DataGen, it is intended for distilling capabilities into open-source large language models through fine-tuning.

Use Cases

• Fine-tune an open-source LLM on the high-reasoning-depth prompts to create a distilled version of Gemini 3 Flash Preview.
• Analyze the reasoning patterns within the 1.01 million tokens of generated synthetic text to understand model behavior.
• Use the dataset's prompts, which include contributions from reedmayhew, for benchmarking reasoning performance in language models.

Strengths

• Contains 1.01 million tokens of generated text, providing substantial material for model training.
• Explicitly designed with a high reasoning depth setting, targeting complex reasoning tasks.
• Generated using a specific model version (Gemini 3 Flash Preview) via the DataGen tool, ensuring a consistent source.

Limitations

• The dataset is entirely synthetic, which may not reflect real-world reasoning patterns or user queries.
• No column definitions, sample data, or row count are provided, making structural assessment and direct usage difficult.
• The source and composition of prompts are only partially known, with some from a named contributor and the rest unspecified.

Provenance

Source

TeichAI via DataGen.

Collection Method

Synthetic generation using the Gemini 3 Flash Preview model with high reasoning depth.

Freshness

Last updated on December 19, 2025.