Superior-Reasoning-SFT-gpt-oss-120b: 435K Samples for Long Chain-of-Thought Training

Chaiyphop created this open-source dataset containing 435,000 samples for training Long Chain-of-Thought models. It is described as a high-quality collection constructed using a principled method, rather than random sampling or heuristic filtering. The dataset was last updated on January 29, 2026.

Use Cases

• Fine-tuning language models for improved reasoning based on the described high-quality, principled sample selection.
• Training models on long-form reasoning chains based on the dataset's focus on Long Chain-of-Thought (Long-CoT) training.
• Benchmarking model reasoning performance against a curated set of 435,000 instruction-response pairs.
• Democratizing access to high-performance training data for reasoning models, as suggested by the dataset's open-source nature.

Strengths

• Contains 435,000 samples, providing a substantial base for model training.
• Designed specifically for training high-performance Long Chain-of-Thought (Long-CoT) models.
• Constructed using a described principled method, contrasting with standard distilled datasets.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-01-29 14:07:23; freshness should be verified.

Provenance

Source

huggingface

Collection Method

Constructed using a described principled method for sample selection.

Freshness

2026-01-29