AI Potluck
Model components / Fine-tuned / chat models

StarCoder2

BigCode Project

BigCode's self-aligned instruct release of StarCoder2 at 16B, trained with a fully transparent pipeline and evaluated on EvalPlus, LiveCodeBench and DS-1000. The base StarCoder2-15B it builds on was trained on The Stack v2 and belongs to the base_pretrained shelf.

Scored on the instruct SKU; the declared artifact is the base repository, so the computed adoption figure reads the base rather than the instruct SKU, which lands at the same level either way. Verified 2026-08-13 via both model cards.

Openness

4 high confidence
4.0
weights
open
data
open(self-oss-instruct-sc2-exec-filter-50k + 7 documented intermediate datasets)
code
open(starcoder2-self-align pipeline on GitHub)
base-data
open(The Stack v2)
license
BigCode-OpenRAIL-M(open but RAIL behavioral-use restrictions, not OSI)

The most open artifact in this category by evidence: the weights, the full training corpus (The Stack v2) and the self-align pipeline are all published. Held to 4 rather than 5 because BigCode-OpenRAIL-M is not OSI-approved - its behavioral-use clauses discriminate against fields of endeavor, which the Open Source Definition forbids - and the top score is reserved for an OSI license. The class reads open weights rather than open source because a 4 cannot be paired with the open-source class. Those RAIL clauses restrict conduct rather than commerce, which is what keeps a fully open pipeline off the restricted floor it would otherwise hit on the license alone.

  • https://huggingface.co/bigcode/starcoder2-15b-instruct-v0.1 recorded 2026-08-13

    "the very first entirely self-aligned code Large Language Model (LLM) trained with a fully permissive and transparent pipeline"; Code bigcode-project/starcoder2-self-align and Dataset bigcode/self-oss-instruct-sc2-exec-filter-50k named on the card, plus a Full Data Pipeline section; `license:bigcode-openrail-m` in the repo tags and `"gated":false`

  • https://huggingface.co/bigcode/starcoder2-15b recorded 2026-08-13

    base StarCoder2 trained on the open The Stack v2; `license:bigcode-openrail-m`; `"gated":false`

Adoption

1 medium confidence
1.0

Research-significant - a fully open self-alignment recipe - but with low production reach. The instruct SKU reads 856 Hugging Face downloads in the trailing 30 days and the base StarCoder2-15B reads 4,425, which is level 1 (<10K) on either SKU and on their sum.

Capability

3 high confidence
3.0

Strong HumanEval for a fully-open 16B in 2024 (best fully-self-aligned code model at release), but HumanEval is saturated and the model is below 2026 frontier coders on harder agentic benchmarks (SWE-bench). Mid-tier.

Verified 2026-08-13