Hunter Alpha Xhigh: 140,000 Reasoning Traces for Math and Coding

140,000 reasoning traces distilled from the Hunter Alpha model via OpenRouter, published by ianncity in March 2026. The data spans 500 million tokens and focuses on high-complexity reasoning across math, coding, and science domains.

Use Cases

• Fine-tuning LLMs for mathematical reasoning using the 30% Algebra, Calculus, and Probability traces
• Improving multi-language code generation using the 30% coding distribution covering Rust, Lua, and C#
• Training scientific reasoning models using the 15% physics, chemistry, and biology traces

Strengths

• 140,000 reasoning traces
• 500 million tokens as of March 2026
• Multi-language coding coverage including Rust, Lua, Ruby, and C#

Limitations

• Synthetic data distilled from an AI model which may inherit source model hallucinations
• Domain imbalance with creative writing representing only 10% of the total records

Provenance

Source

Hunter Alpha model on OpenRouter

Collection Method

Synthetic distillation

Time Range

Up to March 2026

Freshness

Last updated March 14, 2026; author indicates ongoing updates until the source model is removed from OpenRouter.