StarCoder is an open code-generation model from the BigCode community, trained on permissively licensed code from GitHub in more than 80 programming languages.
In plain words
BigCode is an open scientific collaboration focused on building code models responsibly. StarCoder, from May 2023, has 15.5 billion parameters, an 8,000-token context and can fill in missing code in the middle of a file. It was trained on 1 trillion tokens from The Stack, a dataset of permissively licensed repositories whose authors could opt out. StarCoder2, released later, came in 3B, 7B and 15B sizes, with training data four times larger.
A quick example
In an editor, you write a function name and a comment such as "# return the average of a list, ignoring None values". StarCoder can suggest the function body, or fill a gap between code you already wrote above and below.
Where you'll meet it
- Code completion in editors.
- Research on open, responsibly sourced code models.
Related
Sources
- PaperStarCoder: may the source be with you!, Li et al. (arXiv) · read 29 Sept 2026
- Docsbigcode/starcoder, BigCode (Hugging Face) · read 29 Sept 2026
- PaperStarCoder 2 and The Stack v2: The Next Generation, Lozhkov et al. (arXiv) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.