Code Leaderboard

Code generation and understanding

Models

Benchmarks

Providers

Open Weight

Deployable Now

7 self-host

Structured benchmark-backed rows

Signal-backed rows without full normalized tables

Benchmark-expected rows still catching up

This leaderboard still shows tracked models beyond the ones with full benchmark rows. Use the benchmark badge on each row to tell whether the model is Structured, Provider-reported, Arena only, or still Pending benchmark coverage.

LLMs Image Gen Vision Multimodal

#	Model	Value	Quality	Params	MATH Lvl 5	MMLU	Arena-Hard-Auto	GAIA	$/M In	Speed	Open
6	Qwen3 Coder 30B A3B Instruct Qwen Open Weights