Stable-Code 3B: benchmark results
Stability AI's Stable Code 3B, a decoder-only model pretrained on 1.3 trillion tokens across 18 programming languages, with fill-in-middle and 16K context. Provider: Stability AI. Released 2024-01-16. Access: Open.
Aggregate Rating 1182 ± 24, rank #1583 of 1640 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 19.56 | Accuracy (%) (5-shot) | 40.7 |
| ShaderMatch | 9.64 | Clone Match Rate (%) | 28.6 |
| EvalPlus (HumanEval+ & MBPP+) | 35.7 | Pass@1 avg (%) | 25 |
| EvoEval Tool Use | 41 | Pass@1 (%) | 24 |
| BigCode Models Leaderboard | 30.7 | HumanEval Python Pass@1 (%) | 22 |
| Open LLM Leaderboard v1 - MMLU | 37.53 | Accuracy (%) (5-shot) | 19.2 |
| EvoEval Combine | 3 | Pass@1 (%) | 19 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 41.6 | MC2 (%) (0-shot) | 17.6 |
| EvoEval Difficult | 10 | Pass@1 (%) | 17 |
| EvoEval Concise | 31.1 | Pass@1 (%) | 13 |
| Open LLM Leaderboard v1 - ARC Challenge | 35.32 | Normalized accuracy (%) (25-shot) | 13 |
| Open LLM Leaderboard v1 - HellaSwag | 57.45 | Normalized accuracy (%) (10-shot) | 12.9 |
Interactive version: theaggregate.ai/model?slug=stable-code-3b · How It Works · Data refreshed daily, snapshot 2026-11-07.