LLM Leaderboard

Convex is a platform that is designed to work well with AI coding agents. Compare the performance of different language models writing Convex code and evaluate which models work best for you.

About these evaluations

These scores reflect performance on seven categories of benchmarks designed to test language model capabilities across different Convex coding patterns. The benchmarks can be found on GitHub.

Correctness

Measures ability to write Convex code that follows a specification.

Efficiency

Tests how optimal the generated code is, which includes using indexes, args/returns validation, and other optimizations.

Understanding

Evaluates ability to understand existing code, like database schemas, and add new functionality.

Instruction Following

Tests ability to accurately follow complex, multi-step instructions.

Ready to build your own AI-powered application?

Build with Convex + AI →