Claude Mythos Distilled 25K: Synthetic SFT Dataset for LLM Training

25,000 high-quality synthetic examples designed for supervised fine-tuning of language models. The dataset was created by WithinUsAI and is formatted as JSONL with chat messages and metadata. It was last updated on May 18, 2026.

Use Cases

• Fine-tuning language models for cybersecurity and red teaming tasks based on the described category.
• Training models to exhibit agentic behavior based on the described reasoning style.
• Developing models with technical depth for specialized domains based on the balanced categories mentioned.
• Creating instruction-following assistants based on the user/assistant chat message pairs.

Strengths

• Contains 25,000 high-quality examples.
• Balanced across general and specialized capability categories, including cybersecurity.
• Includes rich metadata alongside the chat message pairs.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is synthetic, which may introduce artifacts not present in real-world data.

Provenance

Source

WithinUsAI

Collection Method

Synthetic generation, likely distilled from a frontier model.

Freshness

Last updated 2026-05-18 00:45:03; freshness should be verified.