DeepSeek V4.1 Flash 1017: try it online
Ask DeepSeek's new Flash model a question, or show it an image, right here. No sign-up; a few free answers a day.
Tested and updated October 18, 2026
We tested it: three real examples
Each example below went through the try box above, with the model's answer copied as it came back. Times are from a request in the US to the answer arriving.
Asked: Turn this receipt into a table with item, quantity and price, then check that the items add up to the total.
| Item | Quantity | Price |
|---|---|---|
| Organic Bananas | 2.1 lb | 1.66 |
| Oat Milk | 64oz | 4.49 |
| Sourdough Loaf | 1 | 5.25 |
| Eggs Large | 12ct | 3.99 |
| Cheddar Block | 8oz x2 | 7.18 |
Our check: every line and the sum are right (4.2 s). One quirk: it put pack sizes like "64oz" in the quantity column; ask for "size" and "quantity" as separate columns if you need them apart.
Asked: What does this chart show? Give the trend, the biggest jump and anything unusual, in three short bullet points.
- Overall trend: Monthly active users show a strong upward trend, growing from 35,000 in February to 145,000 in September 2026.
- Biggest jump: The largest month-over-month increase occurred between June and July, where users grew by 40,000 (from 60,000 to 100,000).
- Unusual point: There is a noticeable dip in June, where active users dropped from 75,000 in May down to 60,000 before resuming growth.
Our check: all three points are correct, including the units and the size of the jump (3.2 s).
Asked: Why does this fail, and how do I fix it? Show the corrected line.
csv.DictReader returns all values as strings. When sum() starts with the default integer 0 and tries to add a string from the "amount" column, Python raises TypeError.
total = sum(float(row["amount"]) for row in rows)Our check: right cause, working fix (3.7 s). It read the code and the comment straight from the screenshot.
What it is
DeepSeek-V4.1-Flash-1017 is the October 17, 2026 update of DeepSeek's V4.1 Flash model. DeepSeek's announcement points to stronger benchmark performance and faster work on tasks, and the open weights went to No. 1 on Hugging Face's trending models.
It takes text and images in and gives text out. Hugging Face lists it as image-text-to-text.
What you can do with it
- Read a screenshot or a photo of a document and get the text, a table or a summary back.
- Explain a chart, a diagram or an error dialog.
- Write and fix code, including from a screenshot of code.
- Everyday writing: summaries, emails, outlines, translations.
What it can't do: it does not generate images, audio or video.
Model facts
| Full name | DeepSeek-V4.1-Flash-1017 |
|---|---|
| Maker | DeepSeek |
| Input / output | Text and images in, text out |
| Weights | deepseek-ai/DeepSeek-V4.1-Flash-1017 on Hugging Face |
| Context length | 1M tokens |
| Modes | Thinking and non-thinking (thinking is on by default in DeepSeek's API) |
| Model name in DeepSeek's API | deepseek-flash |
| OpenRouter model id | deepseek/deepseek-v4.1-flash-1017 |
Already on deepseek-v4-flash? DeepSeek retired that model and deepseek-v4-flash-vision-exp: requests under those names are now answered by V4.1 Flash 1017 and billed at the Flash price.
DeepSeek V4.1 Flash pricing: where to call it
US dollars per million tokens, checked October 18, 2026. Cheaper isn't always better: providers differ in speed, uptime and context length.
| Where | Input | Output |
|---|---|---|
| DeepSeek API, off-peak | $0.15 | $0.60 |
| DeepSeek API, peak | $0.30 | $1.20 |
| OpenRouter via DeepInfra | $0.078 | $0.235 |
| OpenRouter via Relace | $0.05 | $0.60 |
| OpenRouter via Cloudflare | $0.10 | $0.60 |
| OpenRouter via DeepSeek | $0.15 | $0.60 |
| OpenRouter via Fireworks | $0.30 | $1.20 |
DeepSeek's peak hours are 01:00–04:00 and 06:00–10:00 UTC on weekdays, which is the US evening (6–9 pm and 11 pm–3 am PDT). Cached input on DeepSeek's API costs $0.003 off-peak. OpenRouter adds no markup to model prices but charges 5.5% when you buy credits by card. Sources: DeepSeek pricing, OpenRouter.
DeepSeek V4.1 Flash API: use it from code
DeepSeek's own API is OpenAI-compatible. Turn thinking off for quick answers:
curl https://api.deepseek.com/chat/completions \
-H "Authorization: Bearer $DEEPSEEK_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "deepseek-flash",
"thinking": {"type": "disabled"},
"messages": [{"role": "user", "content": "Hello!"}]}'Or through OpenRouter, with one key for many models:
curl https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "deepseek/deepseek-v4.1-flash-1017",
"messages": [{"role": "user", "content": "Hello!"}]}'Questions
Can DeepSeek V4.1 Flash 1017 generate images?
No. It reads images and text and answers in text. It can describe a photo, read a screenshot or a chart, but it does not draw.
Is it free to try here?
Yes: a few answers a day, no sign-up. For more, use DeepSeek's own app or the API.
What does "1017" mean?
The release date, October 17. DeepSeek has named updates this way before (DeepSeek-V3-0324, DeepSeek-R1-0528), so a later update gets a new number.
How much does the API cost?
On DeepSeek's API, $0.15 per million input tokens and $0.60 per million output tokens off-peak, double that at peak hours. Through OpenRouter it starts at $0.05 input (Relace) or $0.078 input and $0.235 output (DeepInfra). Prices checked October 18, 2026; see the table above.
Why did my short request come back empty?
Thinking is on by default in DeepSeek's API, and the reasoning counts toward your token limit. With a small max_tokens the answer can be empty. Send "thinking": {"type": "disabled"} or raise the limit.
How long a document can it read?
Its context is 1M tokens, enough for several long books in one request.