Unverified50% confidenceFactExact time
Qwen3于2025年4月发布，Dense模型引入QK RMS Norm，MoE模型取消共享专家并改用更细粒度专家切分配全局批量负载均衡损失，训练数据翻倍到36万亿Token
1
Sources
50%
Confidence
Long-term
Relevance
7/22/2026
First Seen
Sources
Related Claims
UnverifiedQwen2于2024年6月发布，将GQA铺到全系列，KV缓存节省40%，训练数据从3万亿Token增长到7万亿72% similarUnverifiedThe Qwen 3 series was released in 2025 and employs Mixture of Experts (MoE) architecture and Chain-of-Thought enhancement techniques.71% similarUnverifiedThe Qwen3 series is a family of open-source models released by Alibaba in 2025 that excels in code generation and reasoning tasks; the programming-specific Qwen3 loads at approximately 20 GB and can expand to 35–40 GB during inference.70% similarVerifiedQwen3 is a new generation open-source large model series released by Alibaba Cloud in 2025, supporting a hybrid reasoning mode.68% similarUnverifiedThe Qwen 2.5 series was launched in 2024, covering a full range of open-source weights from 0.5B to 72B.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/585440API
curl https://kongchang.com/api/v1/knowledge/claims/585440MCP
get_claim(id=585440)