FINAL-Bench/Darwin-28B-REASON

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 17, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Cold

Darwin-28B-REASON is a 27.6 billion parameter model developed by FINAL-Bench, derived from Darwin-28B-Opus and enhanced for advanced reasoning. It utilizes Reasoning-Trace Distillation (RTD) and the proprietary Darwin-DELPHI test-time engine to achieve 89.39% on the GPQA Diamond benchmark. This full standalone model, based on Qwen3.6 generation architecture, supports a 262,144 token context length and excels at graduate-level scientific and mathematical reasoning tasks.

Loading preview...

Overview

Darwin-28B-REASON is a 27.6 billion parameter model from FINAL-Bench, specifically engineered for advanced reasoning. It is a standalone model, meaning it requires no external base or adapter, and is derived from the Darwin-28B-Opus base model. The model integrates two core components:

  1. Reasoning-Trace Distillation (RTD): This process distills complete reasoning chains into the model, enhancing its ability to handle long-form, multi-step scientific reasoning while retaining bilingual capabilities.
  2. Darwin-DELPHI: A proprietary test-time reasoning engine that employs multi-sample cross-validation, re-examination of uncertain responses, and iterative self-critique to converge on a consensus answer. This engine is not stored in the model weights but is crucial for its top-tier performance.

Together, these technologies enable Darwin-28B-REASON to achieve 89.39% on the GPQA Diamond benchmark, a PhD-level graduate science reasoning evaluation. The model features a Qwen3.6 generation architecture, supports a 262,144 token context length, and is released under the Apache 2.0 license.

Key Capabilities

  • Exceptional performance in graduate-level STEM reasoning and mathematical problem-solving.
  • Handles complex multi-step chain-of-thought tasks effectively.
  • Supports code generation and debugging.
  • Strong bilingual reasoning in English and Korean, with secondary support for Chinese and Japanese.

Recommended Use Cases

  • Graduate-level STEM reasoning (e.g., GPQA, science qualifying exams).
  • Mathematical problem solving (e.g., MATH, AIME-style problems).
  • Complex multi-step chain-of-thought tasks.
  • Code generation and debugging.
  • Bilingual reasoning applications requiring strong English and Korean support.

Finetunes
1 models