DeepSeek V4.1 Flash: What It Is, API Price and Whether to Use It (2026)

DeepSeek V4.1 Flash is DeepSeek's low-cost "Flash" model, formally released on September 10, 2026, priced at RMB 1 per million cache-miss input tokens and RMB 4 per million output tokens off-peak (RMB 0.02 per million for cache-hit input). Since September 14, 2026 at noon Beijing time, requests sent to DeepSeek V4 Pro have been routed to V4.1 Flash and billed at Flash-series rates, so if you call V4 Pro through DeepSeek's API you are already using it. Updated October 26, 2026.

DeepSeek V4.1 Flash at a glance

	DeepSeek V4.1 Flash
Maker	DeepSeek
Formal release	September 10, 2026 (TechNode, citing a notice sent to API users)
Off-peak price, cache-hit input	RMB 0.02 per million tokens
Off-peak price, cache-miss input	RMB 1 per million tokens
Off-peak price, output	RMB 4 per million tokens
Peak-hour price	Twice the off-peak price
Context window	1,048,576 tokens (OpenRouter listing)
Inputs	Text and images; output is text (OpenRouter listing)
Architecture	Sparse mixture-of-experts, 552B-parameter backbone (OpenRouter listing)
V4 Pro routing	From September 14, 2026, 12:00 Beijing time

Prices are in RMB because that is how DeepSeek's notice was reported; TechNode cites a Chinese-language IT Home report for them. Peak-hour prices are stated as twice the off-peak prices; the implied peak rates are RMB 0.04 / 2 / 8. We could not open DeepSeek's own pricing page, so check it before budgeting, and treat dollar figures on third-party gateways as their own conversions.

What changed from V4 Pro

• Routing: per TechNode, DeepSeek said that from September 14 requests to V4 Pro are routed to V4.1 Flash. If your code names V4 Pro, your bill and your outputs now come from the Flash model.
• Performance claim: TechNode reports that DeepSeek said V4.1 Flash "had surpassed V4 Pro in performance, cost, speed and total completion time" in internal and external testing. This is DeepSeek's claim as reported; we did not find an independent benchmark table.
• Vision: the OpenRouter listing shows image input, which headlines in launch coverage call "native vision". Check that your provider enables it before you rely on it.
• A variant ID: gateway listings also show deepseek-v4.1-flash-1017. We found no official description of it, so treat it as an unverified, separately dated build, and pin the model ID you test.

Should you use DeepSeek V4.1 Flash?

Use it when cost per token matters more than the last few points of quality: high-volume classification and extraction, summarising long documents inside a 1M-token window, first-pass code review, and agent loops where most calls are cheap reads. Test before you move critical work: send 20 of your real prompts to V4.1 Flash and to the model you use now, and compare answers and total tokens, because cheaper tokens do not help if the model needs longer outputs.

For hard, open-ended engineering, compare against frontier models first; see Sonnet 5.5 vs Opus 5.5 and GPT-6.1 Sol vs GPT-6 Astra for how current flagship prices look.

How to call it

1. Create a DeepSeek API key, or use a gateway that lists the model (for example OpenRouter).
2. Set the model ID to the Flash model your provider lists (deepseek-v4.1-flash on OpenRouter) and keep your existing OpenAI-style client code.
3. Reuse long prompt prefixes: cache-hit input is priced at RMB 0.02 against RMB 1 for a cache miss, a 50x difference, so stable system prompts and documents cut the bill most.
4. Avoid peak hours for batch jobs, since peak prices are double.

FAQ

What is DeepSeek V4.1 Flash? It is DeepSeek's Flash-series language model, formally released on September 10, 2026, with a roughly 1M-token context window and image input according to the OpenRouter listing.

How much does the DeepSeek V4.1 Flash API cost? Off-peak it is RMB 0.02 per million cache-hit input tokens, RMB 1 per million cache-miss input tokens and RMB 4 per million output tokens; peak-hour prices are double. Confirm the current numbers on DeepSeek's pricing page.

Is DeepSeek V4.1 Flash better than V4 Pro? DeepSeek says it surpassed V4 Pro in performance, cost, speed and total completion time, and since September 14 V4 Pro requests are routed to it. That is the vendor's claim, not an independent benchmark.

What is deepseek-v4.1-flash-1017? It is a model ID that appears in gateway listings. We found no official explanation, so we cannot say how it differs.

Is this site affiliated with DeepSeek? No. GeneratorNest is independent; DeepSeek is named only to describe the model.