Expired60% confidenceFactTime unknown
DeepSeek V4 chose not to use N-gram technology, abandoning the Multi-token Prediction combined with N-gram approach that had been explored in V3/R1.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
6/3/2026
First Seen
Valid until: 9/1/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedDeepSeek previously explored Multi-token Prediction (MTP) combined with N-gram in V3/R179% similarUnverifiedDeepSeek V4 Pro的参数规模远小于Kimi K3或Qwen3.8这类模型72% similarUnverifiedHY3 has not yet displaced the open-source benchmark GLM 5.2, but trades wins and losses with DeepSeek V4 Pro on programming and agent benchmarks.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/39927API
curl https://kongchang.com/api/v1/knowledge/claims/39927MCP
get_claim(id=39927)