deepseek-v4.1IMAGE

DeepSeek V4.1 Flash 1017 scores 82.2% on DualBench100, below the older version's 84.2%

Citing the DualBench100 accounting benchmark, the user notes V4.1 Flash 1017 scores only 82.2% versus 84.2% for the older version, while Gemini 4 Argon hits 91.1% and GLM-5.3 Flash 88.1%.

丰收

@ai_fengshouArchived 10 days agoOriginally posted 2026.10.22

DeepSeek V4.1 Flash 1017 scores 82.2% on DualBench100, below the older version's 84.2%

丰收 @ai_fengshouOriginally posted 2026.10.22

Open original on X

新版不如旧版，这升级升了个寂寞 DeepSeek V4.1 Flash 1017，会计基准 DualBench100 只拿 82.2%，旧版 84.2% 榜首 Gemini 4 Argon 91.1%，开放最强 GLM-5.3 Flash 88.1% 我部署模型前一定先看榜单，版本号真不是越大越强 https://www.businesswire.com/news/home/20261021130097/en/DeepSeeks-Latest-Update-Scores-Lower-on-Accounting-Work-as-Open-Models-Trail-on-Consistency-DualEntry-Benchmark-Finds

x.com/ai_fengshou/status/2113105813753430075Archived 2026.10.22 11:59