Contamination MapText / LLM / Model
Phi-4-mini-flash-reasoning
- Organisation
- Microsoft
- Family
- Phi · version 4-mini-flash-reasoning
- Released
- 2025-06-19 (HF repo creation of microsoft/Phi-4-mini-flash-reasoning (2025-06-19); card: 'Release date: June 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-flash-reasoning
- Weights
- Open · MIT
- Links
- PaperModel card
Training data disclosure: Vague
Level determined for 6% of benchmark cells.
- TestTest data seen0
- SeenSeen, split unclear0
- TrainTraining data seen0
- Related data seen0
- NoNot seen1
- ?Unknown14
“We pre-train our 3.8B-parameter model Phi4-mini-Flash with 5T tokens from the same Phi4-mini data corpus”
Other declared training data
-
Phi-4-mini pre-training corpus (5T tokens) (broad web data: may contain benchmark test sets) · split: unknown
Same corpus as phi-4-mini (filtered public documents + synthetic).
Evidence
“We pre-train our 3.8B-parameter model Phi4-mini-Flash with 5T tokens from the same Phi4-mini data corpus”
arxiv.org · paper · Section 1 · retrieved 2026-10-15 -
synthetic math reasoning data generated by DeepSeek-R1 (incl. public math questions) (not yet matched to a dataset) · split: unknown · fine-tuning
Evidence
“The training data for Phi-4-mini-flash-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 “a curated selection of high-quality, publicly available math questions and a part of the SFT(Supervised Fine-Tuning) data that was used to train the base Phi-4-mini-flash model;”
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
New SambaY architecture pre-trained from scratch on the Phi-4-mini corpus, then distilled following the Phi-4-mini-reasoning recipe.
Sources
- Decoder-Hybrid-Decoder Architecture for Efficient Reasoning with Long Generation · paper · retrieved 2026-10-15
- Phi-4-mini-flash-reasoning model card · model card · retrieved 2026-10-15