CodeGemma
Designed for real projects, it produces structured outputs (e.g., JSON patches, diffs) and works with tool/function calling so agents can run linters, unit tests, or compilers in the loop. It’s optimized for efficiency—quantization and compact variants make it viable on a single modern GPU or edge devices—yet it retains strong reasoning for nontrivial tasks across popular languages (Python, JS/TS, Java, Go, C/C++, Rust, Shell, and more).
Typical uses include IDE autocompletion, repo-level refactoring, code review assistants, RAG over internal codebases, and CI bots that propose fixes with explanations. If you’re building a coding copilot or automated code agent and want open-weight flexibility with solid speed-to-quality tradeoffs, CodeGemma is a great fit.
Overview
CodeGemma is Google’s code-specialized Gemma variant, tuned for software tasks like code generation, completion (incl. fill-in-the-middle), refactoring, debugging, docstring and test creation. It’s efficient, runs on a single GPU or locally with quantization, and supports tool/function calling and structured (JSON) outputs.
About Google
At Google, we think that AI can meaningfully improve people's lives and that the biggest impact will come when everyone can access it.
View Company Profile