StarCoderBase
The original 15.5B open code model — BigCode's 2023 landmark trained on The Stack.
About
StarCoderBase (May 2023) is BigCode's original 15.5B code model: the first major open code LLM, trained on The Stack (6T tokens, 86 languages) with the opt-out governance that defined responsible open code training.
Licensed under BigCode OpenRAIL-M: not OSI open source — use restrictions apply. Historically foundational (the base for StarCoder, OctoCoder, and dozens of fine-tunes), currently obsolete for new builds: 2023 quality, 8K context, superseded many times over. 1.6k Hugging Face downloads on the canonical checkpoint. Included as the landmark that proved the BigCode governance model — for current work, see starcoder2-15b (already in the catalog) or Qwen2.5-Coder.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-claimed at release (2023): leading open HumanEval — treat as provider-claimed and dated. Superseded many times over.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 5/10
Scored 5/10 (metrics tier). The landmark that launched open code LLMs and proved the opt-out governance model. Docked five points: (1) OpenRAIL-M is not open source; (2) quality is 2023-vintage and obsolete — modern 7B models beat it; (3) 8K context; (4) minimal current use (1.6k HF downloads); (5) no active maintenance. A historical entry; fine-tune from StarCoder2 or Qwen instead.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.2/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 5.3/10
- Manual review · sanity read + red-flag check
- 4.0/10
- · HF downloads 1,592 (peer-normalized)
- · released 1258 days ago
- · manual review: landmark that launched open code LLMs; 2023 vintage, quality obsolete; license: BigCode OpenRAIL-M — not OSI; use restrictions apply