Allen InstituteLanguage model
OLMo 2 32B
OLMo 2 32B has no overall index value yet because there are too few independent benchmark results.
PopularityRank 548 of 785
29out of 100-1 vs. last month
based on search, usage and attention
Summary
OLMo 2 32B is the largest family member in Allen AI’s OLMo 2 release. Its model card describes 6 trillion training tokens, 32 billion parameters, and the same deliberately short 4,096-token context as OLMo 2 13B. Transparency is part of the release. Training data, code, checkpoints, and methods remain available for research and inspection.
The 0325 release is its own training point, not merely an older 13B model expanded to 32B. The larger weight set adds local capacity but keeps the 4,096-token context. OLMo 2 32B is a research checkpoint, not a chat variant.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 18 published benchmark results
Page 1 of 2
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
OLMES MMLU (Show measurement, test conditions, and source)
- Source value
- 74.9
- Score
- 74.9
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Test; 14,042 items; 5-shot; char normalization; MCF only.
OLMES ARC-Challenge (Show measurement, test conditions, and source)
- Source value
- 90.4
- Score
- 90.4
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Test; 1,172 items; 5-shot; PMI normalization.
OLMES HellaSwag (Show measurement, test conditions, and source)
- Source value
- 89.7
- Score
- 89.7
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Validation; 1,000 sampled from 10,042; 5-shot; char normalization.
OLMES WinoGrande (Show measurement, test conditions, and source)
- Source value
- 83
- Score
- 83
- Metric
- accuracy
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Validation; 1,267 items; 5-shot; no normalization.
OLMES Natural Questions (Show measurement, test conditions, and source)
- Source value
- 50.2
- Score
- 50.2
- Metric
- F1
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Validation; 1,000 sampled from 3,610; 5-shot.
OLMES DROP (Show measurement, test conditions, and source)
- Source value
- 74.3
- Score
- 74.3
- Metric
- F1
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Validation; 1,000 sampled from 9,536; 5-shot.
OLMES AGIEval English (Show measurement, test conditions, and source)
- Source value
- 61
- Score
- 61
- Metric
- MCF accuracy
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Test; 2,646 items; 1-shot; held-out.
OLMES GSM8K (Show measurement, test conditions, and source)
- Source value
- 78.8
- Score
- 78.8
- Metric
- exact match
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Test; 1,319 items; 8-shot CoT; held-out.
OLMES MMLU-Pro (Show measurement, test conditions, and source)
- Source value
- 46.9
- Score
- 46.9
- Metric
- MCF accuracy
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Test; 12,032 items; 5-shot; held-out.
OLMES TriviaQA (Show measurement, test conditions, and source)
- Source value
- 88
- Score
- 88
- Metric
- F1
- Unit
- %
- Benchmark version
- OLMES Table 6 / Appendix A.1 (report v3 2025-10-08)
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Allen Institute for AI OLMo 2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. OLMo 2 32B base-model row in Table 6 of report v3 (2025-10-08), evaluated with OLMES. Report defines multiple-choice scores as accuracy and generative scores as F1. Appendix A.1 supplies splits and shots. Exact base checkpoint only; no Instruct results transferred.; Validation; 7,993 items; 5-shot; held-out.
Independent Clinical Evaluation of General-Purpose LLM Responses to Signals of Suicide Risk (Show measurement, test conditions, and source)
- Source value
- 14
- Score
- 14
- Metric
- Invitation to continue, overall sample average
- Unit
- %
- Benchmark version
- arXiv:2510.27521v1
- Category
- clinical-safety
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Judd et al., Independent Clinical Evaluation of General-Purpose LLM Responses to Signals of Suicide Risk
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact model named by the paper: OLMo-2-32b as system under test, served through a vLLM API on the open-source Dyff AI assessment framework in private Kubernetes. Forty-three US-based practicing therapists and trainees role-played seven-turn risk-factor sequences and annotated responses with a clinician-developed codebook. The paper reports 829 annotated responses overall; this is not assigned as the N for each model-estimated probability.
- Context
- Study-specific annotated response probabilities, not clinical outcomes or a broad ranking. The paper gives no per-estimate denominator, so sampleCount is omitted. Reported 95% intervals are retained. The paper notes that the 20% versus 5% invitation contrast is not significant at conventional levels (odds ratio 0.229, p≈0.08).
Independent Clinical Evaluation of General-Purpose LLM Responses to Signals of Suicide Risk (Show measurement, test conditions, and source)
- Source value
- 20
- Score
- 20
- Metric
- Invitation to continue, hopelessness prompt
- Unit
- %
- Benchmark version
- arXiv:2510.27521v1
- Category
- clinical-safety
- Direction
- Higher is better
- Lower confidence bound
- 7.8
- Upper confidence bound
- 42.1
- Source type
- independent-evaluator
- Evaluator
- Judd et al., Independent Clinical Evaluation of General-Purpose LLM Responses to Signals of Suicide Risk
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Exact model named by the paper: OLMo-2-32b as system under test, served through a vLLM API on the open-source Dyff AI assessment framework in private Kubernetes. Forty-three US-based practicing therapists and trainees role-played seven-turn risk-factor sequences and annotated responses with a clinician-developed codebook. The paper reports 829 annotated responses overall; this is not assigned as the N for each model-estimated probability.
- Context
- Study-specific annotated response probabilities, not clinical outcomes or a broad ranking. The paper gives no per-estimate denominator, so sampleCount is omitted. Reported 95% intervals are retained. The paper notes that the 20% versus 5% invitation contrast is not significant at conventional levels (odds ratio 0.229, p≈0.08).
Page 1 of 2
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
OLMo 2 32BERNIE 4.5 300B A47B 29Sonar Deep Research 29Tiny Aya Earth 29Nemotron Nano 2 VL 12B 29
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as OLMo 2 32B, chosen by model line, developer, size, price and release date.
- OLMo 2 13BAllen Institute · same developer
- QwQ 32BAlibaba · similar size
- DeepSeek-R1-Distill-Qwen-32BDeepSeek · similar size
- Qwen 3 30B A3BAlibaba · similar size
- Aya Vision 32BCohere · similar size
- Gemma 3 27BGoogle · similar size
- Qwen 3 Coder 30B A3BAlibaba · similar size
- Aya Expanse 32BCohere · similar size
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about OLMo 2 32B
How large is the context window of OLMo 2 32B?
4,096 tokens. That is how much text OLMo 2 32B reads at once, about 10 A4 pages.
Page count as a rule of thumb from Artificial Analysis
GLM-5.1200K
DeepSeek-V3 0324164K
GLM-4.5131K
GPT OSS 120B131K
DeepSeek-V3.2128K
DeepSeek-R1128K
OLMo 2 32BRank 103 of 1034K
Can I run OLMo 2 32B locally?
Yes, the weights are open.
Stated by Allen Institute
- Open weights
- yes
- License
- Apache 2.0
- Parameters
- 32B
- active per token
- n/a
Sources and data date
Show all 11 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.