BigCode·released 2024-02-221 source
StarCoder 2 3B benchmark scores: 4 benchmarks tracked. 0% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 4 benchmarks·best result 21.6 on GSM8K (self-reported)·0 of 4 independently reproduced
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| GSM8K | 21.6accuracy (%) | self-reported· optimizedT1 | 2024-02-22Qwen2.5-Coder Technical Report |
| MMLU | 15.47accuracy (%) | self-reported· optimizedT1 | 2024-02-22Qwen2.5-Coder Technical Report |
| Winogrande | 14.2accuracy (%) | self-reported· optimizedT1 | 2024-02-22Qwen2.5-Coder Technical Report |
| ARC AI2 | 12.27accuracy (%) | self-reported· optimizedT1 | 2024-02-22Qwen2.5-Coder Technical Report |
Claims drawn from cited facts, not live model generation.
This model is a compact 3 billion parameter model developed by BigCode in 2024, a scale that favors efficiency and lower deployment cost over the extra headroom of larger systems.
3 cited facts
This model is tracked across 4 benchmarks, but it currently holds no top score on any of them. Because these results are self-reported and not yet independently confirmed, they should be read as claims rather than verified results.
3 cited facts
With an average SOTA gap of 71.18 points on the disclosed harness, this model trails the benchmark leaders by a wide margin and sits well behind the front of the pack; it also holds the top score on zero SOTA leaderboards, so it currently shows no category leadership.
2 cited facts
2 facts cross-checked across data sources: 2 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: curated_param_counts, epoch_benchmarks
StarCoder 2 3B is an AI model developed by BigCode, released 2024-02-22. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
StarCoder 2 3B has recorded scores on 4 benchmarks, each shown with its evidence status.
StarCoder 2 3B has recorded scores on 4 benchmarks — Winogrande, MMLU, ARC AI2, GSM8K. The full table above shows each score with its evidence status.
0 of 4 recorded scores (0%) are independently reproduced rather than self-reported by the lab.
StarCoder 2 3B has 4 tracked claims: 4 self-reported.