DeepSeek V4.1 Flash 1017 is now available on JZS Token at a rate that is 54.5% below the official list price. That number is the same for input tokens, output tokens, and cache hits, and it holds during the standard weekday peak windows because both rates double there.
This article shows the exact comparison, how the math works, what the peak windows mean for a real workload, and how to switch to the new model ID without changing anything else in your integration.
The short answer
The official list price for this model is $0.15 per million input tokens, $0.60 per million output tokens, and $0.003 per million cached input tokens. JZS Token charges $0.06825, $0.273, and $0.001365 respectively. In every segment, our price is 45.5% of the official rate, which means the saving is 54.5%.
Those are standard, non-batch prices, checked on November 11, 2026. They are not introductory coupons and they do not require a minimum spend. The live numbers are always on the model page: jzstoken.com/model-pricing/deepseek-v4.1-flash-1017.
Official price vs JZS Token
| Billing item | Official list price | JZS Token | You save |
|---|---|---|---|
| Input, per 1M tokens | $0.15 | $0.06825 | 54.5% |
| Output, per 1M tokens | $0.60 | $0.273 | 54.5% |
| Cache hit, per 1M tokens | $0.003 | $0.001365 | 54.5% |
| Peak input, per 1M tokens | $0.30 | $0.1365 | 54.5% |
| Peak output, per 1M tokens | $1.20 | $0.546 | 54.5% |
| Peak cache hit, per 1M tokens | $0.006 | $0.00273 | 54.5% |
The percentage is not an average. Each individual line is exactly 54.5% lower, so the saving does not shrink or grow depending on whether your workload is input-heavy or output-heavy.
The math behind 54.5%
For input tokens, $0.06825 divided by $0.15 is 0.455. The same ratio appears for output tokens ($0.273 / $0.60 = 0.455) and for cache hits ($0.001365 / $0.003 = 0.455).
Because the ratio is the same in all three segments, a blended 3:1 input/output mix also lands at the same discount. The official blended rate is (3 × $0.15 + $0.60) / 4 = $0.2625 per million tokens. The JZS blended rate is (3 × $0.06825 + $0.273) / 4 = $0.1194375. That is 54.5% lower.
The absolute saving grows with volume. For 10 million input tokens and 10 million output tokens, the official bill would be $1.50 + $6.00 = $7.50. At the JZS rates, the same usage costs $0.6825 + $2.73 = $3.4125, a saving of $4.0875. At 100 million tokens on a 3:1 blend, the difference is about $14.31.
Peak and off-peak pricing
This model uses time-of-day pricing on both sides. The standard peak windows are Monday through Friday, 09:00–12:00 and 14:00–18:00 Beijing time. Outside those windows, including all day Saturday and Sunday, the off-peak rates above apply.
During the standard weekday peak windows, both the official rate and the JZS rate double. The input rate becomes $0.30 official versus $0.1365 on JZS; output becomes $1.20 versus $0.546; cache hits become $0.006 versus $0.00273. The ratio is still 0.455, so the 54.5% saving still holds.
The window is determined by when the request starts, not by how long the response takes. A request that starts at 11:59 Beijing time is a peak request even if the last token arrives at 12:02.
One practical caveat: official holiday schedules can add exceptions that are not part of the normal weekday pattern. If you are scheduling a large batch around a Chinese public holiday, check the live model page before you start. For ordinary weekdays, weekends, and the windows above, the comparison in this article is the right one.
What the price looks like for common workloads
A coding assistant that sends 8 million input tokens and receives 2 million output tokens in a month is a common shape. At official rates, that is $1.20 input + $1.20 output = $2.40. At JZS rates, it is $0.546 input + $0.546 output = $1.092. The monthly difference is $1.308 for that one workload.
A document pipeline that reads a large corpus and produces short summaries has a higher input-to-output ratio. Ten million input tokens and one million output tokens cost $1.50 + $0.60 = $2.10 officially, versus $0.6825 + $0.273 = $0.9555 on JZS. The saving is $1.1445.
These examples use list prices and do not include cache hits, retries, or failed requests. They are meant to show the direction and scale of the difference, not to predict a specific invoice.
Cache pricing is part of the comparison
Cache hits are billed at $0.001365 per million tokens on JZS, versus $0.003 officially. That is the same 54.5% reduction. If your application sends a long, stable system prompt or repeatedly references the same document, cache hits can be a meaningful part of the bill, so the comparison should include them rather than looking only at input and output.
For this model, there is no separate cache-write rate on our side. Cache-write tokens are billed at the input rate. The live model page always shows the rates that billing will use.
Why a gateway can be cheaper than the official list
JZS Token is not the model vendor. We buy capacity through contracted infrastructure and route requests through it. That supply arrangement is why we can publish a rate below the official list while keeping one stable API, one balance, and one invoice for many models.
This is different from a marketplace where the price changes depending on which provider happens to serve a request. Our published rate is the rate we charge. A request that is served successfully is billed at the numbers on the model page; there is no auction at request time.
It also means the comparison is not “official vendor vs the same vendor’s other checkout.” It is the official list price versus a reseller’s contracted rate. That is exactly what the table above compares, line by line.
What the model is good for
DeepSeek V4.1 Flash 1017 has a 1 million token context window and accepts text input. It is a good fit for long codebases, large documents, multi-file review, extraction, summarization, and agent loops where the context is large but the required output is relatively short.
It does not accept images. If your workflow depends on screenshots, charts, or scanned pages, use a vision-capable model for that step and route the text-only steps here.
We do not publish a synthetic “best for everything” benchmark. Test the model on your own prompts, especially if you are replacing a pro or reasoning tier. A cheaper model that needs extra retries is not cheaper in production.
How to call it
The model is available through the same OpenAI-compatible endpoint as the rest of the catalog. Use the model ID exactly as written:
curl https://jzstoken.com/api/v1/chat/completions \
-H "Authorization: Bearer $JZS_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4.1-flash-1017",
"messages": [{"role": "user", "content": "Summarize this document."}],
"stream": false
}'The base URL is https://jzstoken.com/api/v1. The request shape is the same as any OpenAI-compatible chat completion, so existing SDKs work without special code.
Migrating from the previous DeepSeek models
The previous DeepSeek IDs in our catalog have been retired: deepseek-v4-flash, deepseek-v4-pro, and deepseek-v4.1-flash. Requests to those IDs now return a model-not-available error.
If you were using one of them, change only the model string to deepseek-v4.1-flash-1017. The API key, base URL, request body, and SDK configuration stay the same. If you had pinned a pro or reasoning tier for a specific workflow, test this model on that workflow first; a flash model is cheaper, but it is not automatically a capability-for-capability replacement.
Billing details to keep in mind
JZS Token bills per token in USD, against one account balance shared by every API key on the account. The number shown on the pricing page is the number billing reads. If a rate changes, the live page is the authority; this article is a snapshot of the rates above, checked on November 11, 2026.
Requests that fail at the upstream never produce a successful charge. A request that is accepted and returns a response is settled from the usage reported by the upstream. Cache hits, cache writes, input, and output are recorded separately so you can reconcile a bill against your own token counters.
If you are building an agent that makes many small calls, the output allowance and the retry policy matter more than the headline input rate. Set a small max_tokens for test calls, watch the usage page, and scale up after you confirm the model behaves well on your data.
FAQ
Is the 54.5% saving available on every request?
It applies to the standard off-peak rates and to the standard weekday peak windows shown above, where both sides double. Holiday schedules can introduce exceptions, so check the live rate before a large scheduled batch.
Do I need a new API key?
No. Use your existing key. Only the model ID changes.
Is there a batch discount on top?
No. The rates in this article are the standard non-batch rates. We do not apply an additional batch multiplier on top of them.
Can I use the model with Claude Code, Cline, or Cursor?
Yes, through the OpenAI-compatible chat completions endpoint. Use the model ID above and the base URL https://jzstoken.com/api/v1. Claude-specific clients that require the Anthropic Messages endpoint should use an Anthropic-compatible model instead.
What happens if the official price changes?
We review our catalog when upstream pricing changes. The model page is updated with the current rate, and the price there is the one billing uses. Do not assume this article’s snapshot remains current forever.
Try it on your own workload
Create an account, use your existing key, and send one small request with the model ID deepseek-v4.1-flash-1017. If the response is what you expect, compare your real token counts and cache behavior against your current provider before moving a larger workload.
Create a free JZS Token account or view the live model pricing.