DeepSeek has released V4.1 Flash, a small model that understands images and beats the larger V4 Pro in tests, so clearly that DeepSeek is retiring the Pro model. On ChatX, V4.1 Flash is free to use, guests included. The key questions in short form.
What is new in V4.1 Flash?
DeepSeek V4.1 Flash is the first model of a new architecture family. It uses only a small part of its parameters per request, which makes it fast and cheap. For long conversations it needs a fraction of the memory of the previous generation. DeepSeek passed those savings on as a price cut on 10 September. Built-in image understanding is new as well. A dedicated image encoder was trained together with the language model from the start, where DeepSeek had only offered this as an experiment before.
Why is V4 Pro disappearing?
DeepSeek writes that tests by several parties put V4.1 Flash ahead of V4 Pro on quality, cost, speed and total runtime. Since 14 September all requests to V4 Pro have been routed to V4.1 Flash, and the Pro model is being retired. On ChatX there is therefore only one DeepSeek entry left, V4.1 Flash. What made V4 Pro special, the thorough thinking before an answer, now sits as a switch inside the same model. It is unusual for a provider to replace its more expensive model with a cheaper one instead of offering both side by side.
What can it do with images?
The documentation names three cases: describing images, reading text from images, reading charts. In practice on ChatX that means:
- Screenshots: photograph an error message and ask what it means, upload part of a spreadsheet and have the values written out as text.
- Charts: upload a bar or line chart and have the trend explained or the numbers estimated.
- Photos: translate menus, signs or forms, or have a photo described to get alt text.
Handwriting and very small type are the limit. Gemini 3.8 Flash or Sol are more reliable there.
How does it differ from the other free model?
ChatX has two free models, GPT-5 nano and DeepSeek V4.1 Flash. The difference is clear:
- GPT-5 nano: text only, very fast, good for short questions, translations and simple rewrites. No images.
- DeepSeek V4.1 Flash: understands images, is noticeably stronger at programming and analysis and writes better structured long answers. Slightly slower in return.
If you start out as a guest, DeepSeek is the better pick for almost everything. Only on very short questions does nano finish sooner.
What should you watch out for?
- Data protection: DeepSeek answers run through servers in China. For confidential content that belongs in your choice of model.
- Thinking is off by default: V4.1 Flash answers straight away. With a subscription or purchased tokens you can set “Depth of thought” to “High”. The model then reasons before answering, which helps noticeably with calculations, code and multi-step questions. Guests and users without credit stay on the fast direct mode.
- Guest limit: as a guest, answer length is capped. For long texts or code it is worth registering. The model stays free either way.
V4.1 Flash shows where the market is heading. Small models are catching up with the large ones, and image understanding is becoming standard, free models included.