By No Code MBA · Facts checked
- Provider
- DeepSeek
- Announcement date
- Status at last check
- Generally available
- Model family
- DeepSeek V4
- Preceding model
- DeepSeek V4 Flash
- Capabilities
- Text, Image, Coding
What changed
V4.1 Flash is the smallest model in DeepSeek's new architecture family and adds native image understanding, replacing V4 Flash and V4 Flash Vision Exp. It is a 552B-parameter MoE with 8B active parameters for input and 16B for output, a context of up to 1M tokens, and KV cache needs DeepSeek puts at about a quarter of the previous generation's HBM. DeepSeek reports 90.9 on GPQA Diamond and 74.2 on DeepSWE v1.1, and an October 17 update (V4.1-Flash-1017) raised DeepSWE to 77.0.
Access and availability
Use the model name deepseek-flash on the DeepSeek API, which is now served by V4.1-Flash-1017; legacy deepseek-v4-flash names route to it. Weights are on Hugging Face at deepseek-ai/DeepSeek-V4.1-Flash under the MIT License, and it also runs in DeepSeek's web and app.
Official access documentationWhat you can build
Test screenshots and document images alongside the text-only tasks in your evaluation set.
Limitations
DeepSeek says open-source inference support is still being worked out with the community. Its own tables show it trailing larger frontier models on some tests, such as HLE without tools. Benchmark claims are DeepSeek's own.