DeepSeek V4.1 Pro: release status, and how V4.1 Flash differs from V4
The DeepSeek V4.1 generation is shipping in stages — the small model first, the flagship later. Here is what is out, what DeepSeek has promised, and what it means if you use DeepSeek today.
What’s known so far
- DeepSeek V4.1 Pro was announced by DeepSeek itself; no release date, size or pricing yet.
- DeepSeek V4.1 Flash shipped on 10 September 2026: 552B parameters, 8B active when reading input and 16B when generating output.
- V4.1 Flash is DeepSeek’s first model to understand images natively, with no separate vision version.
- V4.1 Flash has open weights.
- DeepSeek V4 Pro stays on: DeepSeek planned to retire it, then changed course.
At a glance
| Detail | Value |
|---|---|
| Models in the generation | DeepSeek V4.1 Pro · DeepSeek V4.1 Flash |
| V4.1 Pro | Announced, no release date |
| V4.1 Flash | Released 10 September 2026, open weights |
| V4.1 Flash parameters | 552B; 8B active on input, 16B on output |
| Images | Native, starting with V4.1 Flash |
| Live in Twelver now | DeepSeek V4 Pro · DeepSeek V4 Flash |
When is DeepSeek V4.1 Pro coming out?
There is no date. In the V4.1 Flash announcement DeepSeek said a temporary arrangement for older models would last “until V4.1-Pro launches” — the first time the company named the model officially. Size, context, pricing and benchmarks are all still unpublished. As soon as V4.1 Pro is available we’ll connect it and update this page.
What V4.1 Flash already does
V4.1 Flash is the smallest model on the new architecture. Only a small share of its parameters runs per request, so it is fast, and it needs several times less memory for long context than the previous generation. It reads images on its own, with no separate vision variant. DeepSeek says third-party tests put V4.1 Flash ahead of V4 Pro on quality, cost and speed. If that holds, V4.1 Pro should be a real step up from today’s flagship.
What you can use today
Twelver runs DeepSeek V4 Pro and V4 Flash, both with a 1M-token context and a thinking mode you can switch off. Connect DeepSeek through our API today and moving to V4.1 takes no rework — only the model answering changes. How to choose between Pro and Flash is in the DeepSeek chapter.
Frequently asked questions
When will DeepSeek V4.1 Pro be released?
How does DeepSeek V4.1 Flash differ from V4 Flash?
Is DeepSeek V4 Pro being retired?
Can I use DeepSeek right now?
While you wait — the current version is live
Open in Twelver chat