Claude Opus 4.6/4.7 Reasoning 8.7K is a synthetic instruction-tuned dataset. The data was generated by Claude Opus 4.6 and Claude Opus 4.7, aiming to provide diverse textual data for training large language models in reasoning, instruction following, and domain-specific tasks.

The dataset is in JSON format, covering various task scenarios such as coding, mathematics, role-playing, and scientific Q&A. It is suitable for training and improving the reasoning and Q&A capabilities of large models.

Dataset Composition

The dataset primarily includes the following four files:

• full_train.jsonl: 8,706 samples, covering all 28 categories;
• instruct_train.jsonl: 7,217 samples, covering 24 instruction-based tasks;
• roleplay_train.jsonl: 1,489 samples, including creative tasks such as heroes, villains, cross-work characters, and narrative writing;
• code_train.jsonl: 1,840 samples, primarily covering programming and mathematics tasks.

Each sample primarily contains the following fields:

• messages: Multi-turn dialogue content in OpenAI Chat format, including system, user, and assistant messages;
• category: The task category to which the data belongs;
• model: The version of the Claude model used to generate the sample.