Contamination MapText / LLM / Model
Phi-4-mini-reasoning
- Organisation
- Microsoft
- Family
- Phi · version 4-mini-reasoning
- Released
- 2025-04-30 (arXiv v1 submission 2025-04-30 (HF repo created 2025-04-29; card: 'Release date: April 2025'))
- Training cutoff
- 2025-02-01
Evidence
“This is a static model trained on offline datasets with the cutoff date of February 2025 for publicly available data.”
huggingface.co · model card · Status · retrieved 2026-10-15 - Checkpoints
- microsoft/Phi-4-mini-reasoning
- Weights
- Open · MIT
- Links
- PaperModel card
Training data disclosure: Partial list
Level determined for 6% of benchmark cells.
- TestTest data seen0
- SeenSeen, split unclear0
- TrainTraining data seen0
- Related data seen0
- NoNot seen1
- ?Unknown14
“we aggregate multiple public datasets—such as Bespoke (Labs, 2025), Openthoughts (OpenThoughts, 2025), and OpenR1-Math (HuggingFace, 2025)—along with several in-house seed datasets.”
Other declared training data
-
AquaRAT (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“AquaRAT (Ling et al., 2017)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
Ape210K (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“Ape210K (Zhao et al., 2020)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
MetaMathQA (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“MetaMathQA (Yu et al., 2023)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
MathInstruct (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“MathInstruct (Yue et al., 2023)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
TAL-SCQ5K (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“TAL-SCQ5K (TAL-SCQ5K, 2023)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
OpenR1-Math (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“OpenR1-Math (HuggingFace, 2025)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
Bespoke-Stratos-17k (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“Bespoke-Stratos-17k (Labs, 2025)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
OpenThoughts-114K (not yet matched to a dataset) · split: unknown · fine-tuning
Used as question/trajectory source for the distillation (mid-training/SFT/DPO) data.
Evidence
“OpenThoughts-114K (OpenThoughts, 2025)”
arxiv.org · paper · Table 2 (data resources for the reasoning dataset) · retrieved 2026-10-15 -
in-house seed datasets with DeepSeek-R1 generated CoT (not yet matched to a dataset) · split: unknown · fine-tuning
Evidence
“we aggregate multiple public datasets—such as Bespoke (Labs, 2025), Openthoughts (OpenThoughts, 2025), and OpenR1-Math (HuggingFace, 2025)—along with several in-house seed datasets.”
arxiv.org · paper · Section 4 · retrieved 2026-10-15 “The training data for Phi-4-mini-reasoning consists exclusively of synthetic mathematical content generated by a stronger and more advanced reasoning model, Deepseek-R1.”
huggingface.co · model card · Training Datasets · retrieved 2026-10-15
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 (2025-02-01)
Evidence (3)
“This is a static model trained on offline datasets with the cutoff date of February 2025 for publicly available data.”
“2025 AIME I problems and solutions. The test was held on Thursday, February 6, 2025.”
“This year the AIME I will take place on February 6, 2025, while the alternate AIME II will take place on February 12, 2025.”
Unknown MATH-500 · training data not disclosed in enough detail
Notes
Continued from Phi-4-mini (pre-training: see phi-4-mini); ~150B tokens of reasoning training. About 10M rollouts over 1.6M samples. Named sources are partial (in-house seeds unnamed). No decontamination statement found.
Sources
- Phi-4-Mini-Reasoning: Exploring the Limits of Small Reasoning Language Models in Math · paper · retrieved 2026-10-15
- Phi-4-mini-reasoning model card · model card · retrieved 2026-10-15