opensynapselabs/arche3.5-codium-0.5B

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:0.5BQuant:BF16Ctx Length:32kPublished:May 16, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

Arche-Codium-500M by Open Synapse Labs is a compact, instruction-finetuned code generation model built on Qwen2.5-Coder-0.5B-Instruct, featuring 0.5 billion parameters and a 32K context length. It is optimized for fast local code completion, achieving an 80% pass rate on HumanEval. This model is designed for edge-friendly deployment, running efficiently on CPUs, MPS, or low-VRAM GPUs for generating executable Python code.

Loading preview...

Arche-Codium-500M: Compact Code Generation

Arche-Codium-500M is a 0.5 billion parameter, instruction-finetuned model developed by Open Synapse Labs, specifically designed for code generation. Built upon the Qwen2.5-Coder-0.5B-Instruct base, it prioritizes efficient local execution and resource-friendly operation.

Key Capabilities & Features

  • Compact Size: With 500M parameters, it runs effectively on CPUs, Apple Silicon (MPS), or low-VRAM GPUs (requiring as little as 2 GB VRAM/RAM).
  • Code Generation Performance: Achieves an 80% pass rate on the HumanEval benchmark (16/20 tasks), demonstrating its capability in generating functional code.
  • Apache 2.0 License: Fully open and commercially usable.
  • CLI Integration: Ready for use with the arche-code CLI for streamlined development workflows.
  • Edge-Friendly: Optimized for deployment on laptops, Raspberry Pi, and mobile devices.

Use Cases & Limitations

This model excels at generating executable Python code, particularly for functions under 50 lines. It is an ideal entry point into the Arche ecosystem for developers needing fast, local code completion with minimal hardware requirements. However, it is not intended as a replacement for larger 7B+ models for complex architectural tasks or as a general chat model. It may struggle with multi-step reasoning, occasionally truncate output, or hallucinate imports, requiring verification of generated code.