Gemini 3.5 Flash Distilled: 25,000 Synthetic Samples for Agentic Reasoning

25,000 synthetic samples designed to replicate the capabilities of the Gemini 3.5 Flash model. The dataset was created by WithinUsAI via template-based evolutionary synthesis with content-normalized SHA-256 deduplication and was last updated on May 20, 2026.

Use Cases

• Fine-tuning language models for agentic execution based on the described frontier-level capabilities.
• Benchmarking rapid multi-step reasoning in AI systems based on the dataset's stated purpose.
• Training models for dense context analysis tasks as indicated in the description.
• Developing advanced autonomous coding assistants based on the optimized, synthetic examples.

Strengths

• Contains 25,000 samples, providing a substantial synthetic corpus.
• Explicitly designed for low-latency inference, suggesting optimization for practical deployment.
• Created with SHA-256 deduplication, which likely reduces redundancy.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known but sample data and file formats are unavailable, limiting pre-download assessment.
• Data is synthetic, which may not fully capture the complexity and distribution of real-world tasks.

Provenance

Source

WithinUsAI via Hugging Face.

Collection Method

Template-based evolutionary synthesis with content-normalized SHA-256 deduplication.

Time Range

null

Freshness

Last updated 2026-05-20 01:24:55; freshness should be verified.

Geography

null