DeepSeek V4.1 Flash 1017 (Max)

DeepSeek says it beats the previous V4.1-Flash on all 11 benchmarks it reported, by 2.0 to 11.1 points, and reports 77.0 on DeepSWE v1.1. The model is a 552B total parameter open-weight reasoning model (16B active, MoE architecture) with a 1M-token context window that accepts text and images. Output reaches about 230 tokens per second on DeepSeek's API, but the model is verbose: it generated 210M tokens on the Intelligence Index, against a median of 140M. The release also changed routing. Deprecated V4 Flash model names are temporarily routed to V4.1 Flash, and the API calling method and pricing remain unchanged. Peak input pricing is $0.30 per million input tokens and $1.20 per million output tokens, with off-peak rates half that. These benchmark figures come from DeepSeek, and none has been independently reproduced, and commenters on Hacker News questioned the speed and pricing methodology. For a team building an AI product, the cost per agent run is the practical question. One comparison estimates a run that consumes 2M input tokens and 200K output tokens costs $0.42 off-peak on V4.1 Flash versus $1.72 for the same run on V4 Pro off-peak pricing. The MIT license allows commercial use and self-hosting, but hardware requirements for the 1017 weights are not disclosed, so self-hosting is an infrastructure decision. Run your own agent evaluations before moving production traffic.