OLMo 2 32B: benchmark results
Allen AI's OLMo 2 32B (dated 0325), the base pretrained checkpoint in their fully open OLMo 2 line, with Instruct, DPO and SFT post-trained siblings built on top of it. Provider: Allen AI. Released 2025-03-13. Access: Open.
Aggregate Rating 1359 ± 52, rank #1662 of 2169 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| XCR-Bench - CSI Identification | 57.71 | Soft F1 (%; normalised Levenshtein similarity of predicted a | 42.9 |
| AA IFBench | 38.1 | Accuracy (%) | 33.7 |
| AI for Education SEND | 62.39 | Accuracy (%) | 16.9 |
| AI for Education Pedagogy - Maths | 55.56 | Accuracy (%) | 16.2 |
| AA MMLU-Pro | 51.06 | Accuracy (%) | 15.2 |
| AI for Education Pedagogy - Social studies | 57.27 | Accuracy (%) | 15 |
| XCR-Bench - CSI Prediction | 31.87 | Exact Match (x100; case-insensitive exact matches of the mas | 14.3 |
| Artificial Analysis Intelligence Index | 5.97 | Intelligence Index | 14 |
| AI for Education Pedagogy | 58.51 | Accuracy (%) | 13.5 |
| AI for Education Pedagogy - Primary | 61.03 | Accuracy (%) | 12.9 |
| AI for Education Pedagogy - Secondary | 56.6 | Accuracy (%) | 12.7 |
| AI for Education Pedagogy - Science | 56.28 | Accuracy (%) | 12.3 |
Interactive version: theaggregate.ai/model?slug=olmo-2-32b · How It Works · Data refreshed daily, snapshot 2026-10-23.