Unverified50% confidenceFactExact time
Qwen4, built on a next-generation architecture, has already entered training.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/7/2026
First Seen
Valid until: 1/5/2027
Sources
Related Entities
Related Claims
UnverifiedThe application has a built-in qwen3 4b model to enable out-of-the-box functionality74% similarUnverifiedQwen3.8-Flash-Next natively supports MTP, indicating that this capability was built in during the model training phase.73% similarUnverifiedQwen3 Next 125B uses a MoE (Mixture of Experts) architecture and has more than twenty thousand expert modules.72% similarUnverifiedQwen2.5系列在后训练阶段引入了大量工具调用对齐数据，并通过RLHF和DPO强化了结构化输出的可靠性72% similarUnverifiedThe poster used Qwen-Next 3.8 for complex task processing and treated it as a reliable benchmark72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/985975API
curl https://kongchang.com/api/v1/knowledge/claims/985975MCP
get_claim(id=985975)