FINAL-Bench/Darwin-28B-REASON

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 17, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

FINAL-Bench/Darwin-28B-REASON is a 27.6 billion parameter reasoning-enhanced standalone model derived from Darwin-28B-Opus, developed by FINAL-Bench. It utilizes Reasoning-Trace Distillation (RTD) and is optimized for graduate-level scientific reasoning, achieving 89.39% on GPQA Diamond with the proprietary Darwin-DELPHI test-time engine. This model excels in complex multi-step chain-of-thought tasks and mathematical problem-solving, supporting a 262,144 token context length.

Loading preview...

Darwin-28B-REASON: Enhanced Scientific Reasoning Model

Darwin-28B-REASON is a 27.6 billion parameter standalone model from FINAL-Bench, built upon the Darwin-28B-Opus base. It integrates two core components: Reasoning-Trace Distillation (RTD), which distills complete reasoning chains into the model, and the proprietary Darwin-DELPHI test-time engine. This combination significantly enhances its performance in graduate-level scientific reasoning, achieving an 89.39% accuracy on the GPQA Diamond benchmark when used with Darwin-DELPHI.

Key Capabilities

  • Advanced Reasoning: Excels in long-form, multi-step scientific and mathematical reasoning tasks.
  • High Performance: Achieves top-tier results on benchmarks like GPQA Diamond, demonstrating strong problem-solving abilities.
  • Standalone Operation: A full self-contained model, requiring no external base models or adapters.
  • Bilingual Support: Strengthened for English and Korean, with secondary support for Chinese and Japanese.
  • Extended Context: Supports a substantial context length of 262,144 tokens for long-chain reasoning.

Recommended Use Cases

  • Graduate-level STEM reasoning: Ideal for tasks akin to GPQA or science qualifying exams.
  • Mathematical problem solving: Suited for complex problems similar to MATH or AIME-style challenges.
  • Complex multi-step chain-of-thought tasks: Designed to handle intricate reasoning processes.
  • Code generation and debugging: Capable of assisting with programming-related tasks.