DeepSeek-V4.1-Flash-1017 Scores 77% on DeepSWE 1.1

DeepSeek has revamped its latest Flash-class model as DeepSeek-V4.1-Flash-1017, according to Wccftech. The report says the model scored 77% on the DeepSWE 1.1 benchmark.

Wccftech says the score outperformed GLM-5.3, Kimi K3, GPT-5.6 Sol and Opus 5. The source describes the model as tiny but does not provide its size, and it gives no benchmark methodology or comparative scores.

Before the model’s official launch, DeepSeek post-training lead Wu Yu said: “Entering the era of democratized wish-driven coding.” The report does not provide a launch date or further details about the model.