DeepSeek releases V4.1-Flash, an open-weight AI model that can also read images

On September 10, 2026, the AI company DeepSeek released DeepSeek-V4.1-Flash. Its weights, the learned numbers that make a model work, can be downloaded from Hugging Face under the MIT License, which allows commercial use. The model is also available through DeepSeek's API. The model has 552 billion parameters, or learned numbers, in total, but it uses a mixture-of-experts design, in which only part of the model switches on at a time. DeepSeek says just 8 billion parameters are active while reading input and 16 billion while writing output, which helps keep costs low. It understands images as well as text, and VentureBeat reports it can take in up to 1 million tokens, the small pieces of text AI models process, at once. DeepSeek says tests put V4.1-Flash ahead of its V4-Pro model on performance, cost and speed, so starting September 14 it is routing V4-Pro requests to the new model. VentureBeat noted that the benchmark results it reported came from DeepSeek's own evaluations. Why it matters: Open-weight models let students, researchers and small companies download, study and run capable AI themselves instead of only renting it from a few providers. Source: DeepSeek — https://api-docs.deepseek.com/news/news260910/ Source: VentureBeat — https://venturebeat.com/technology/deepseek-v4-1-flash-debuts-with-0-003-1m-off-peak-cached-input-rate-and-benchmarks-eclipsing-gpt-5-6-sol-claude-opus-5