Contamination MapText / LLM / Model
Qwen3
- Organisation
- Alibaba Cloud (Qwen Team)
- Family
- Qwen · version 3
- Released
- 2025-04-29 (Date of the Qwen3 announcement blog post, https://qwenlm.github.io/blog/qwen3/)
- Checkpoints
- Qwen/Qwen3-0.6B, Qwen/Qwen3-1.7B, Qwen/Qwen3-4B, Qwen/Qwen3-8B, Qwen/Qwen3-14B, Qwen/Qwen3-32B, Qwen/Qwen3-30B-A3B, Qwen/Qwen3-235B-A22B, Qwen/Qwen3-0.6B-Base, Qwen/Qwen3-1.7B-Base, Qwen/Qwen3-4B-Base, Qwen/Qwen3-8B-Base, Qwen/Qwen3-14B-Base, Qwen/Qwen3-30B-A3B-Base
- Weights
- Open · Apache 2.0
- Links
- PaperModel cardCode
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
“All Qwen3 models are trained on a large and diverse dataset consisting of 119 languages and dialects, with a total of 36 trillion tokens.”
“all Qwen3 models are publicly accessible under Apache 2.0.”
Other declared training data
-
web data (broad web data: may contain benchmark test sets) · split: unknown
Evidence
“To build this large dataset, we collected data not only from the web but also from PDF-like documents.”
qwenlm.github.io · website · Pre-training · retrieved 2026-10-15 -
PDF-like documents (text extracted with Qwen2.5-VL, refined with Qwen2.5) (broad web data: may contain benchmark test sets) · split: unknown
Provenance of the PDFs is not stated; treated as open-ended (could include papers containing benchmark items).
Evidence
“To further expand the pre-training data corpus, we first employ the Qwen2.5-VL model (Bai et al., 2025) to perform text recognition on a large volume of PDF-like documents.”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 -
synthetic data from Qwen2.5, Qwen2.5-Math and Qwen2.5-Coder (textbooks, QA, instructions, code) (not yet matched to a dataset) · split: unknown
Evidence
“Besides, we employ Qwen2.5 (Yang et al., 2024b), Qwen2.5-Math (Yang et al., 2024c), and Qwen2.5-Coder (Hui et al., 2024) models to synthesize trillions of text tokens in different formats, including textbooks, question-answering, instructions, and code snippets, covering dozens of domains.”
arxiv.org · paper · Section 3.1 · retrieved 2026-10-15 -
long-CoT cold-start dataset (math, code, logical reasoning, STEM; responses from QwQ-32B) (not yet matched to a dataset) · split: unknown · fine-tuning
Sources of the problems are not named.
Evidence
“We begin by curating a comprehensive dataset that spans a wide range of categories, including math, code, logical reasoning, and general STEM problems.”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 “After reserving a validation query set, we generate N candidate responses for each remaining query using QwQ-32B (Qwen Team, 2025).”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 -
reasoning RL query-verifier pairs (3,995) (not yet matched to a dataset) · split: unknown · fine-tuning
Evidence
“We ultimately collect a total of 3,995 query-verifier pairs, and employed GRPO (Shao et al., 2024) to update the model parameters.”
arxiv.org · paper · Section 4.2 · 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
Thinking/non-thinking modes share the same checkpoints. The report states no benchmark decontamination; the cold-start response filter removes responses 'suspected of being overly similar to potential validation set items', which refers to Qwen's own reserved validation query set, not named benchmarks. No training-data cutoff stated. No Qwen3-32B-Base or Qwen3-235B-A22B-Base repo was found on Hugging Face.
Sources
- Qwen3 Technical Report · paper · retrieved 2026-10-15
- Qwen3: Think Deeper, Act Faster · website · retrieved 2026-10-15
- Qwen3-8B model card · model card · retrieved 2026-10-15