droplychee/DropLychee-1.2
DropLychee-1.2 is a 27 billion parameter model developed by droplychee, featuring a 32768 token context length. This model demonstrates strong performance across a diverse set of benchmarks, including high accuracy in multiple-choice tasks like MathArena Aime 2026 (94.1%) and MMLU Pro (86.2%), alongside notable code generation capabilities with a 77.2% resolved rate on SWE Bench Verified. It is particularly distinguished by its balanced proficiency in reasoning, mathematical problem-solving, and code generation, making it suitable for complex analytical and development tasks.
Loading preview...
DropLychee-1.2: A Versatile 27B Parameter Model
DropLychee-1.2 is a 27 billion parameter language model designed for a broad range of tasks, featuring an extensive context length of 32768 tokens. Developed by droplychee, this model exhibits strong performance across various benchmarks, highlighting its capabilities in reasoning, mathematical problem-solving, and code generation.
Key Capabilities & Performance Highlights
- Mathematical Reasoning: Achieves 94.1% accuracy on the MathArena Aime 2026 benchmark and 84.3% on MathArena Hmmt Feb 2026, indicating robust mathematical problem-solving skills.
- General Knowledge & Reasoning: Demonstrates 86.2% accuracy on MMLU Pro and 87.8% on the Diamond multiple-choice task, showcasing strong general knowledge and reasoning abilities.
- Code Generation: Excels in coding tasks with a 53.5% pass@1 rate on SWE Bench Pro and a 77.2% resolved rate on SWE Bench Verified, making it highly effective for software development and bug fixing.
- Multimodal Understanding: Achieves 75.8% accuracy on MMMU Pro Vision, suggesting capabilities in processing and understanding multimodal inputs.
- Text Generation & Instruction Following: Scores 48.2% on SkillsBench V1.1 and 59.3% on Terminalbench 2, indicating proficiency in text generation and following complex instructions.
Ideal Use Cases
DropLychee-1.2 is well-suited for applications requiring a combination of advanced reasoning, precise mathematical computation, and efficient code generation. Its strong performance across diverse benchmarks makes it a versatile choice for:
- Complex Problem Solving: Tackling intricate mathematical and logical challenges.
- Software Development: Assisting with code generation, debugging, and understanding codebases.
- Advanced AI Assistants: Powering intelligent agents that require broad knowledge and reasoning capabilities.
- Research & Development: Serving as a foundational model for further fine-tuning on specialized tasks.