DeepSeek has officially released the DeepSeek-V4.1-Flash model, the smallest entry in a new architecture family designed for higher throughput, faster inference, and expanded scaling potential. This new architecture features native multimodal visual understanding. The V4.1-Flash model demonstrates high-tier performance across several benchmarks, including a GPQA Diamond score of 90.9, a Codeforces rating of 3471, and a 65.6 score on MathArena Apex. It also shows strong capabilities in specialized tasks, such as a 90.6 on Terminal-Bench 2.1 and a 74.2 on DeepSWE v1.1. Following the release, the company retired the previous V4 Flash and V4 Flash Vision Exp models, temporarily routing their legacy names to the V4.1 version. Additionally, DeepSeek has introduced flexible thinking effort controls (low, high, and max) for the V4-Pro and V4-Flash models to allow users to balance task complexity with computational efficiency. API pricing has been adjusted to reflect the new model release, and the company has implemented a peak/off-peak pricing structure to optimize resource allocation. The DeepSeek API also now natively supports the OpenAI Responses API format and is optimized for Codex.