Instruction-Tuning Dataset for AI Reasoning Tasks

2160 examples formatted as multi-turn conversations for training large language models. The dataset was created by Crownelius and last updated in March 2026. It is a formatted version of the original Opus-4.6-Reasoning-2100x dataset.

Use Cases

• Fine-tune language models for instruction-following using the structured 'system', 'user', and 'assistant' message roles.
• Benchmark model reasoning capabilities on the 2160 provided conversational examples.
• Train models to generate structured 'metadata' outputs based on conversational context.
• Use the 'messages' array format to develop or test conversational AI agents.

Strengths

• Contains 2160 formatted conversational examples.
• Data is structured in a consistent JSON format with defined roles.

Limitations

• The sample size of 2160 examples is relatively small for large-scale model training.
• The original source and method of data generation are not detailed, limiting reproducibility.
• Content is synthetic, which may not reflect real-world reasoning patterns.

Provenance

Source

Formatted version of crownelius/Opus-4.6-Reasoning-2100x.

Collection Method

Converted to match the TeichAI/claude-4.5-opus-high-reasoning-250x format.

Freshness

Last updated March 2026.