Code LLM
A large language model trained or fine-tuned predominantly on source code and developer text, optimised for generation, editing and reasoning about programs.
Code LLMs range from small open models specialised for completion (StarCoder, Code Llama, Qwen-Coder, DeepSeek-Coder) to frontier general models whose training mix is heavy on code and which power agents (GPT-5, Claude Opus/Sonnet 4.x, Gemini 2.5/3). Since 2024 the frontier for coding is dominated by general reasoning models with tool use rather than code-only models; leaderboards like SWE-bench Pro and Aider's polyglot benchmark rank them.