Skip to content
TrustList

TrustList summary

StarCoder2 15B comes from the BigCode project and is a code model rather than a general chat assistant. Its Hugging Face model card describes a 15 billion parameter model trained on more than 600 programming languages, drawn from a dataset called The Stack v2, with material excluded wherever owners had asked to opt out.

Built for completing code, not for following orders

The card is direct about how the model should be used. It was trained on GitHub code plus selected extra sources such as Arxiv and Wikipedia, and it is not an instruction-tuned model. A request phrased as a command, such as asking it to write a function that computes a square root, does not work well. The model is better treated as a code continuation engine, which is why the card walks through loading it with the transformers library and generating from a function stub. It also shows how to run it at full precision, in bfloat16, and in 8-bit and 4-bit quantized forms, with the memory footprint falling at each step so that smaller GPUs can host it.

Attention design and context

BigCode reports a context window of 16,384 tokens, combined with a sliding window attention of 4,096 tokens and grouped query attention. Training used the fill-in-the-middle objective, which teaches a model to complete a gap between a prefix and a suffix, and covered over four trillion tokens across one million pretraining steps. The card lists a transformer decoder architecture, bfloat16 precision, 1024 H100 GPUs, the NeMo framework and PyTorch.

Attribution, limits and licence

Because the pretraining data was filtered for permissive licences and unlicensed code only, BigCode still warns that output can reproduce source verbatim, and that the original licence might require attribution. It points to a search index over the pretraining data for tracing where generated code came from. The card adds that code can be inefficient, buggy or exploitable and is not guaranteed to work. The weights are released under the BigCode OpenRAIL-M v1 licence.

Summary written by TrustList from the business's own published information, Nov 2026.

Benchmarks & AI stats

BenchmarkOfficialCommunity avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.

StarCoder2 15B launches

Request a demo or quote from StarCoder2 15B

Protected by reCAPTCHA - Google Privacy Policy and Terms apply.

By sending, you agree we may share your request and contact details with the provider once you confirm your email.

Reviews

Write the first review of StarCoder2 15B

Used it? Your experience helps other buyers decide.

Write a review

Questions & answers

No questions yet. Be the first to ask about StarCoder2 15B.

StarCoder2 15B alternatives

Similar listings buyers compare against this one.

Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems

Cantina Security's first open-weight security model, a post-trained GLM-5.3-Flash for vulnerability research and exploit checks

Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure

Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions

Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms

Liquid AI's open-weight decision model, giving yes or no, choice or score answers from text and images

Aleph Alpha's open-weight English-German mixture-of-experts model for self-hosting, released under Apache 2.0 in October 2026

Microsoft AI's text-to-speech model for keeping one voice across 23 languages, with voice cloning from short audio