davinci-002: benchmark results
Replacement GPT-3 base Completions API model for fine-tuning and legacy completions. Provider: OpenAI. Released 2023-08-22. Access: API.
Aggregate Rating 1331 ± 30, rank #1400 of 1635 rated models, from 35 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MERA Original Paper | 38.3 | Total Score (%; mean of 17 problem-solving and exam tasks, M | 94.4 |
| MERA Original Paper - MathLogicQA | 35.3 | Accuracy (%; 5-shot, log-likelihood) | 94.4 |
| MERA Original Paper - ruMMLU | 61.3 | Accuracy (%; 5-shot, log-likelihood) | 94.4 |
| MERA Original Paper - ruOpenBookQA | 67.5 | Accuracy (%; 5-shot, log-likelihood) | 94.4 |
| MERA Original Paper - ruWorldTree | 76.6 | Accuracy (%; 5-shot, log-likelihood) | 94.4 |
| MERA Original Paper - SimpleAr | 92.7 | Exact Match (%; 5-shot, greedy generation) | 88.9 |
| MERA Original Paper - ruModAr | 47.6 | Exact Match (%; greedy generation) | 88.9 |
| MERA Original Paper - ruMultiAr | 17.6 | Exact Match (%; 5-shot, greedy generation) | 88.9 |
| SAD - Anti-Imitation | 15.39 | Score (%, higher is better) | 80 |
| SAD - Anti-Imitation (Situating Prompt) | 15.39 | Score (%, higher is better) | 80 |
| SAD - Predict Words | 15.47 | Self-prediction score (%, higher is better) | 36.4 |
| SAD - Predict Words (Situating Prompt) | 14.2 | Self-prediction score (%, higher is better) | 31.8 |
Interactive version: theaggregate.ai/model?slug=davinci-002 · How It Works · Data refreshed daily, snapshot 2026-11-05.