New todaydeepseek-v4.1IMAGE
Strata local four-model test: V4.1 Flash best at long context but slow
Switching four local models in Strata on the same task, the tester found DeepSeek V4.1 Flash strongest on long text and instructions but slow, while Qwen3.8 Coder was fastest at code but broke on Japanese.
yutacar@niigata
@yutacarArchived 6 hours agoOriginally posted 2026.10.28
yutacar@niigata @yutacarOriginally posted 2026.10.28
ローカルLLMをStrataで4つ切り替えて、同じ課題で比較しました UD-IQ4_XS：どの課題も崩れない本命 DeepSeek V4.1 Flash：長文と指示は最強、ただし遅い Qwen3.6-35B-A3B：23GBと軽く、日本語もUD並み Qwen3.8 Coder：コードは最速、日本語は崩れる #ローカルLLM #Strata
x.com/yutacar/status/2115192666703196331Archived 2026.10.28 12:05