[AutoBe] Achieved 100% compilation success of backend generation with "qwen3-next-80b-a3b"

AutoBE  is an open-source project enabling automatic generation of backend applications through AI chatbot conversations.

We achieved 100% compilation success for backend applications using local AI models like qwen3-next-80b-a3b — a marked improvement from earlier attempts where most projects failed to build due to compilation errors.

Key Achievement

Rather than having AI generate code as text, AutoBE developed a custom compiler using AST (Abstract Syntax Tree) construction through function calling. This approach presented significant challenges, as not a single model could handle AutoBE’s AST structure during earlier experiments. However, newly released local LLMs overcame this limitation within months.

Important Caveats

100% compilation success doesn’t necessarily mean 100% runtime success. Current limitations include:

• Test function pass rate: approximately 80%
• Occasional SQL query errors
• Potential misinterpretation of complex business logic
• Target goal: 100% runtime success rate by year-end

Project Examples

Results tested across multiple models:

• qwen3-next-80b-a3b-instruct (To Do List, Reddit Community, Economic Discussion, E-Commerce)
• gpt-4.1-mini (same application set)
• gpt-4.1 (same application set)

Future Direction

We plan to develop benchmarks targeting AutoBE compiler components and publish periodic analyses of local LLMs’ function calling capabilities for complex types, with initial benchmarks expected in two months.

• Homepage 
• GitHub