TESSY-SuperGPQA-3K: Programming Contest Training Data for Reasoning Models

CoopReason constructed a 3,000-instance dataset for training reasoning models. The data was synthesized by leveraging GPT-OSS-120B as a teacher model to preserve strong reasoning capabilities while aligning with a student model's distribution. The dataset was last updated on April 17, 2026.

Use Cases

• Fine-tuning language models for programming contest problem-solving based on the described teacher-student cooperation framework.
• Training student models to align with the reasoning distribution of a larger teacher model based on the described synthesis method.
• Benchmarking the reasoning capabilities of language models on synthetic programming data.
• Studying knowledge transfer and data synthesis techniques for model fine-tuning.

Strengths

• Dataset size is explicitly stated as 3,000 instances.
• Designed to preserve the strong reasoning capabilities of the GPT-OSS-120B teacher model.
• Created with a specific methodological framework for teacher-student cooperation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the synthetic generation process and source models.

Provenance

Source

CoopReason

Collection Method

Synthesized by leveraging GPT-OSS-120B as a teacher model within a teacher-student cooperation framework.

Freshness

Last updated 2026-04-17 03:47:43; freshness should be verified.