Qwen 3
#5alibaba · v3 · seit 2025-04-29 · 54× · zuletzt 30. Juni 2026
67
Momentum
Qwen3 is the third generation of Alibaba Cloud's large language model family (Tongyi Qianwen), released on April 28–29, 2025 under the Apache 2.0 license. The series includes eight open-weight models — six dense variants (0.6B–32B parameters) and two Mixture-of-Experts models (Qwen3-30B-A3B and Qwen3-235B-A22B) — as well as the proprietary API model Qwen3-Max with over 1 trillion parameters. All models support a hybrid reasoning mode (thinking/non-thinking) and were trained on approximately 36 trillion tokens covering 119 languages and dialects. The flagship Qwen3-235B-A22B scores 95.6 on ArenaHard, competing with models such as DeepSeek-R1, GPT-o1, and Gemini 2.5 Pro.
Historique du momentum
04.04.03.07.
Fonctionnalités
| Context Window (Tokens) | 32,768 tokens native (Qwen3-32B/8B/etc.), expandable to 131,072 tokens via YaRN; flagship MoE Qwen3-235B-A22B: 262,144 tokens native (Instruct-2507 version) |
| License | Apache 2.0 (for all dense models: 0.6B, 1.7B, 4B, 8B, 14B, 32B as well as MoE models 30B-A3B and 235B-A22B) |
| Platform | Hugging Face, ModelScope, GitHub, Kaggle (download/self-hosting); Alibaba Cloud Model Studio (API); Ollama, LM Studio, llama.cpp, vLLM, SGLang (local inference); chat.qwen.ai (web chat) |
| Price per 1M Tokens | Qwen3-30B-A3B: from $0.08/M input, $0.28/M output; Qwen3-32B: $0.28/M input (via OpenRouter/third-party); Qwen3-Max (API, proprietary): $0.78/M input, $3.90/M output (Alibaba Cloud Model Studio) |
| Release Date | April 28, 2025 (official release of the Qwen3 model family) |
