CodeLlama-7B-Instruct-hf: benchmark results
Meta's CodeLlama 7B Instruct, the instruction-tuned code model in the Llama 2-based CodeLlama family, released under the Llama 2 Community License; the linked card provides no further details here. Provider: Meta. Released 2023-08-24. Access: Open.
Aggregate Rating 1243 ± 12, rank #1545 of 1656 rated models, from 101 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ToolBench - The Cat API | 94 | Task Score | 89.5 |
| ToolBench - VirtualHome | 21.95 | Task Score | 79.1 |
| MMLU-by-task - Econometrics | 32.46 | Accuracy (%) | 77.1 |
| MMLU-by-task - Abstract Algebra | 33 | Accuracy (%) | 75.6 |
| ToolBench - Google Sheets | 41.27 | Task Score | 69.8 |
| Enkrypt AI - Safety Risk | 22.88 | Risk Score | 68.9 |
| ToolBench - Open Weather | 90 | Task Score | 67.4 |
| CIPHER - Neutral Prompts | 71.3 | Vulnerability rate (% of 750 generations with at least one v | 66.7 |
| CIPHER - Secure Prompts | 72.8 | Vulnerability rate (% of 750 generations with at least one v | 66.7 |
| ToolBench Leaderboard | 50.48 | Average Task Score | 65.1 |
| MMLU-by-task - Machine Learning | 34.82 | Accuracy (%) | 60.8 |
| ToolBench - Home Search | 78 | Task Score | 60.5 |
Interactive version: theaggregate.ai/model?slug=codellama-7b-instruct-hf · How It Works · Data refreshed daily, snapshot 2026-10-20.