Contamination MapText / LLM / Model
DeepSeek-V3.2
- Organisation
- DeepSeek-AI
- Family
- DeepSeek · version V3.2
- Released
- 2025-12-01 (DeepSeek API Docs news entry 'DeepSeek-V3.2 Release 2025/12/01', https://api-docs.deepseek.com/news/news1226)
- Checkpoints
- deepseek-ai/DeepSeek-V3.2, deepseek-ai/DeepSeek-V3.2-Speciale
- Weights
- Open · MIT
- Links
- PaperModel card
Training data disclosure: Vague
Level determined for 0% of benchmark cells.
- TestTest data seen0
- SeenSeen, split unclear0
- TrainTraining data seen0
- Related data seen0
- NoNot seen0
- ?Unknown15
“we developed a novel synthesis pipeline that systematically generates training data at scale.”
Other declared training data
-
continued pre-training data aligned with the DeepSeek-V3.1-Terminus 128K long-context extension data (not yet matched to a dataset) · split: unknown
Evidence
“the distribution of training data is totally aligned with the 128K long context extension data used for DeepSeek-V3.1-Terminus.”
huggingface.co · paper · Section 2.1.1 · retrieved 2026-10-15 “Starting from a base checkpoint of DeepSeek-V3.1-Terminus, whose context length has been extended to 128K, we perform continued pre-training followed by post-training”
huggingface.co · paper · Section 2.1.1 · retrieved 2026-10-15 -
GitHub issue-Pull Request pairs (executable software-issue environments for code-agent RL) (broad web data: may contain benchmark test sets) · split: unknown · fine-tuning
Same kind of source as SWE-bench (GitHub issue/PR pairs); no decontamination against SWE-bench is stated.
Evidence
“We constructed large-scale, executable environments for software issue resolution by mining millions of issue-Pull Request (PR) pairs from GitHub.”
huggingface.co · paper · Section 3 (Code Agent) · 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
Unknown AIME 2025 · training data not disclosed in enough detail
Unknown MATH-500 · training data not disclosed in enough detail
Notes
Same continued pre-training as V3.2-Exp; post-training = specialist distillation + mixed RL + large-scale agentic task synthesis. No decontamination statement found in the report.
Sources
- DeepSeek-V3.2 model card · model card · retrieved 2026-10-15
- DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models · paper · retrieved 2026-10-15