Unverified50% confidenceFactExact time
Next-generation models such as DeepSeek-V3 and Qwen3 have natively integrated the MTP mechanism during the pre-training stage.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/5/2026
First Seen
Valid until: 12/4/2026
Sources
Related Entities
Related Claims
UnverifiedMeta在Llama 4论文中将MTP作为核心训练目标，DeepSeek-V3也大规模采用了类似机制79% similarUnverifiedDeepSeek introduced MTP technology as early as version V3.74% similarUnverifiedQwen3's overall performance has pulled ahead of DeepSeek V3 and R1 models by a certain margin.71% similarUnverifiedDeepSeek自2023年成立以来，凭借V2的MoE架构创新和V3的超大规模训练跻身全球顶级模型行列70% similarUnverifiedMTP (Multi-Token Prediction) is a speculative decoding strategy first introduced by DeepSeek in version V3.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/860453API
curl https://kongchang.com/api/v1/knowledge/claims/860453MCP
get_claim(id=860453)