DeepSeek V4-0813 Shows No Gains Over July Version in Benchmarks

teortaxesTex · x · 2026-08-13

A developer reported that DeepSeek's newly released V4-0813 model performs virtually identically to the V4-Flash-0731 version across almost all benchmarks, and even scores lower on SciCode.

Cited discussants speculated that the model is still undertrained, leading to jagged capabilities. Although some aggregate leaderboards show higher scores, the overall model lacks significant breakthroughs.

Related event: DeepSeek V4-0813 Criticized for Showing No Improvement Over Flash Version(2 posts)→

Original post →

More from Models

Models channel →