DeepSeek V4.1 Flash online, no 510 GB download.

What DeepSeek V4.1 Flash is

A multimodal mixture-of-experts model with a 552 billion parameter backbone, released under the MIT licence on 10 September 2026. It activates 8 billion parameters per token while reading your prompt and 16 billion while writing the answer, which is why it is quick for its size. The context window is 1,048,576 tokens and it reasons on every turn.

The engineering behind it is aimed at one thing: the key value cache, the memory a model keeps for every token it has already seen. DeepSeek puts theirs at 890 bytes per token, so a full million token conversation costs roughly 930 MB of cache instead of the tens of gigabytes this usually implies. On long agent runs, where the conversation grows all day, that is the whole game.

The weights are public and permissively licensed, so running it yourself is allowed. It is also a 510 GB checkpoint whose smallest community quantisation on release day was 168.9 GB, with no released runtime able to execute it yet. Hosted is not a compromise here, it is the only route that exists this week.

What you get

• Chat in the browser, with no install, no driver and no VRAM check. Open the Studio, pick the model, type.
• Image input. Paste a screenshot into the composer and ask about it. The older DeepSeek V4 Flash 0731 refuses images; this one reads them.
• Native tool calls, so Agent mode and the coding agent run on it directly with no prompt translation in between.
• A million tokens of context that stays affordable, because the cache behind it is small by design rather than by truncation.
• All 47 chat models on the same account, plus the image, video and audio studios. This is one pick in the list, not a separate product.

Call it from your own code

Create a key under Cloud API keys in account settings, point the base URL at https://lu-labs.ai/api/inference/v1 and send the catalog model id. Anything that speaks the OpenAI chat completions format works, including Aider, LibreChat and a plain curl call. The key spends your credits and cannot read or change your account.

curl https://lu-labs.ai/api/inference/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek-ai/DeepSeek-V4.1-Flash",
       "messages":[{"role":"user","content":"Hello"}]}'

How to start in three steps

1. Buy a €5 credit pack, or take the Hosted plan if you want a monthly budget. Card details go to Stripe and the payment is secured with 3D Secure.
2. The Studio opens on your account. Pick DeepSeek V4.1 Flash in the model picker.
3. Type, or paste a screenshot. For scripts, create an API key in account settings and use the base URL above.

What it costs

The honest limits

This model reasons on every turn and there is no way to switch that off, so a short question still buys a short think. That is a property of the model, not a setting we withheld: we measured the off signal and it made the turn longer, not shorter. The token count above is the floor of what a pack buys, not an average of your week, because input tokens draw from the same wallet and a long conversation carries its whole history into every turn. Credits are one pool for text, images and video together. Cloud runs on shared hardware, so at busy hours a request can wait a few seconds before it starts streaming. The catalog is curated: we host open-weight models we have priced and tested, and you cannot upload your own base model. Account data is hosted in the EU and you can delete it yourself. We never train on your data and we never sell it.

Questions