This page provides the LLM coding benchmark leaderboard, covering SWE-Bench Verified, SWE-Bench Pro, LiveCodeBench, and SWE-bench Multilingual datasets, comparing GPT, Claude, Qwen, and DeepSeek models.
As of 2026-05, this page covers SWE-bench Verified, LiveCodeBench, SWE-Bench Pro - Public, SWE-bench Multilingual and related benchmarks for LLM Coding Benchmark Leaderboard, making it straightforward to compare within the same task family.
Click any model name to check context length, licensing, and pricing on its detail page. See Data Methodology for scoring details.
There is no single, universally accepted coding leaderboard. Static benchmarks like SWE-bench and HumanEval measure specific skills but can be gamed through targeted fine-tuning. We selected two complementary human-preference leaderboards: LMArena Coding Arena ranks models on general programming tasks (debugging, algorithms, code generation) via anonymous crowd-sourced voting; DesignArena Code Category focuses specifically on visual, front-end code generation (websites, UI components, games) using the same blind-voting methodology. Reading both together gives a fuller picture of coding capability.
Elo ratings from anonymous A/B voting on real general coding tasks (debugging, algorithms, code generation) submitted by developers.
Updated 2026-05-14
Elo ratings from anonymous voting on visual front-end code tasks (websites, UI components, games, data viz) by Arcada Labs.
Updated 2026-05-17
Gemma 4 31B
DeepMind
Gemma 4 26B A4B
DeepMind
No qualifying model on this benchmark.
Click any row to open the model page. Tick the checkboxes to compare up to 4 models side by side.
