Contamination Map Methodology

Text / LLM / Model

Olmo 3

Organisation
Allen Institute for AI (Ai2)
Family
OLMo · version 3 (7B: 1025, 32B: 1125)
Released
2025-11-20 (Ai2 blog 'Olmo 3: Charting a path through the model flow to lead open-source AI' dated November 20, 2025 (https://allenai.org/blog/olmo3); arXiv report v1 15 Dec 2025)
Training cutoff
2024-12-01
Evidence
“Date cutoff:** Dec 2024”
huggingface.co · model card · Model Description · retrieved 2026-10-15
“Text extraction We start with 104 dumps from the CommonCrawl corpus, with a cutoff date of December”
arxiv.org · paper · Section 3.4 · retrieved 2026-10-15
Checkpoints
allenai/Olmo-3-1025-7B, allenai/Olmo-3-1125-32B, allenai/Olmo-3-7B-Think-SFT, allenai/Olmo-3-7B-Think-DPO, allenai/Olmo-3-7B-Think, allenai/Olmo-3-32B-Think-SFT, allenai/Olmo-3-32B-Think-DPO, allenai/Olmo-3-32B-Think, allenai/Olmo-3-7B-Instruct-SFT, allenai/Olmo-3-7B-Instruct-DPO, allenai/Olmo-3-7B-Instruct, allenai/Olmo-3-7B-RL-Zero-Math, allenai/Olmo-3-7B-RL-Zero-Code, allenai/Olmo-3-7B-RL-Zero-IF, allenai/Olmo-3-7B-RL-Zero-General, allenai/Olmo-3-7B-RL-Zero-Mix
Weights
Open · apache-2.0
Links
PaperModel cardCode

Training data disclosure: Full list

Level determined for 6% of benchmark cells.

“These models are pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets.”
huggingface.co · model card · Model card intro · retrieved 2026-10-15
“This dataset contains the full mix of documents used to train Olmo 3 7B.”
huggingface.co · dataset card · Source Sizes · retrieved 2026-10-15

Training corpora

Other declared training data

Declared exclusions

Benchmark cells

Classic LLM benchmarks

Unknown MMLU (Measuring Massive Multitask Language Understanding) · training data not disclosed in enough detail

Unknown GSM8K (Grade School Math 8K) · training data not disclosed in enough detail

Unknown MATH (Mathematics Aptitude Test of Heuristics) · training data not disclosed in enough detail

Unknown HumanEval · training data not disclosed in enough detail

Unknown ARC-Challenge (AI2 Reasoning Challenge, Challenge Set) · training data not disclosed in enough detail

Unknown HellaSwag · training data not disclosed in enough detail

Unknown TruthfulQA · training data not disclosed in enough detail

Unknown WinoGrande · training data not disclosed in enough detail

Recent LLM benchmarks

Unknown GPQA Diamond · training data not disclosed in enough detail

Unknown MMLU-Pro · training data not disclosed in enough detail

Unknown LiveCodeBench (code generation, release_v6) · training data not disclosed in enough detail

Unknown SWE-bench Verified · training data not disclosed in enough detail

Unknown AIME 2024 · training data not disclosed in enough detail

Not seen AIME 2025 · its test items were published from 2025-02-06, after the model's stated training-data cutoff (2024-12-01)

Evidence (4)
“Date cutoff:** Dec 2024”
huggingface.co · model card · Model Description · retrieved 2026-10-15
“Text extraction We start with 104 dumps from the CommonCrawl corpus, with a cutoff date of December”
arxiv.org · paper · Section 3.4 · retrieved 2026-10-15
“2025 AIME I problems and solutions. The test was held on Thursday, February 6, 2025.”
web.archive.org · website · retrieved 2026-10-15
“This year the AIME I will take place on February 6, 2025, while the alternate AIME II will take place on February 12, 2025.”
maa.org · website · retrieved 2026-10-15

Unknown MATH-500 · training data not disclosed in enough detail

Notes

7B (Oct 2025) and 32B (Nov 2025) bases were released together on 2025-11-20 and draw on the same Dolma 3 pools with size-specific mixes. Base-model decontamination covers midtraining and long-context data only (any split of OLMES benchmarks, incl. GSM8K, MATH, HumanEval, MMLU, ARC, HellaSwag, WinoGrande per OlmoBaseEval); the 5.9T-token stage-1 web/PDF/code mix is not stated to be decontaminated. Post-training exclusions list the Table 16 benchmarks that map to registry ids (also BBH, ZebraLogic, AGI Eval, HumanEval+, MBPP+, IFEval, IFBench, AlpacaEval, PopQA). Midtraining includes synthetic data generated from the MATH training set (TinyMATH).

Sources