FINAL-Bench/Darwin-28B-Coder
Darwin-28B-Coder is a 28-billion parameter code-specialized language model developed by VIDRAFT FINAL-Bench, built on the Darwin family (Qwen3.5-compatible) architecture with a 32K token context length. It excels in function-level code generation, complex-library composition, and tool/function calling, achieving 100.0% on HumanEval and 72.0% on BigCodeBench-Complete. This model is optimized for high-performance code tasks, competing directly with frontier models like GPT-4o and Claude 3.5 Sonnet.
Loading preview...
Darwin-28B-Coder: A Code-Specialized LLM
Darwin-28B-Coder is a 28-billion parameter language model from VIDRAFT FINAL-Bench, specifically engineered for advanced code generation and understanding. Based on the Darwin family (Qwen3.5-compatible) architecture, it features a substantial 32K token context length, enabling it to handle complex coding tasks.
Key Capabilities & Performance
- Exceptional Code Generation: Achieves a perfect 100.0% on HumanEval, outperforming GPT-4o and Claude 3.5 Sonnet.
- Complex Library Composition: Leads public benchmarks with 72.0% on BigCodeBench-Complete, demonstrating strong capabilities in multi-library code generation.
- Function Calling: Scores 90.0% on Function Calling benchmarks, comparable to Claude 3.7 Sonnet and GPT-4o.
- MBPP Performance: Attains 84.0% on MBPP, showcasing robust problem-solving skills.
- Training Data: Fine-tuned using
m-a-p/CodeFeedback-Filtered-Instruction(Python, AST-validated) for high-quality code output.
Ideal Use Cases
- Function-level code generation: For developers needing precise and efficient code snippets.
- Complex library integration: When working with multiple libraries and requiring sophisticated code composition.
- Tool and function calling: For building agents or systems that interact with external tools and APIs.
- Competitive coding and development: As a powerful assistant for challenging programming tasks.