Flash is right when…
The job is high volume and low reasoning: pulling fields out of documents, sorting items into categories, summarizing, translating, reading images.
DeepSeek V4
Two V4 tiers. One endpoint. Change the model name.
Most workloads do not need a reasoning model — they need a cheap one that is good enough. Knowing which tier you actually need is most of the saving.
The two tiers
| V4.1 Flash | V4.1 Pro | |
|---|---|---|
| Model name to send | deepseek-flash |
deepseek-v4-pro |
| Context window | 1M tokens | 1M tokens |
| Maximum output | 384K tokens | 384K tokens |
| Image input | Yes | No |
| Video input | Yes | No |
| Concurrency | Up to 2,500 | Up to 500 |
| Built for | Volume work: extraction, classification, summarization | Harder reasoning where accuracy matters more than cost |
Notice what is not different: context and output limits are identical. The tiers differ in reasoning depth and in whether the model can see. That makes the choice simpler than it looks.
Choosing between them
Pro is right when…
The answer has to be right and the reasoning is genuinely multi-step — analysis, complex code work, anything where a wrong answer costs more than the tokens.
The mistake worth avoiding
The most common pattern we see is a workload that started on the strongest available model and never got revisited. It worked, nobody had a reason to touch it, and the workload grew tenfold underneath it.
Run a sample of your real traffic through Flash and compare the output against what you get today. If Flash is good enough — and for extraction and classification it usually is — you have found your saving without touching the architecture.
Questions about the models
Are DeepSeek V4 Pro and DeepSeek V4 Flash the same as V4.1?
V4.1 is the current point release of the V4 line, so if you are searching for DeepSeek V4 Pro or DeepSeek V4 Flash, these are the two tiers that answer to those names: V4.1 Pro and V4.1 Flash. When a newer point release ships we will say so on this page rather than quietly renaming things.
Can we use both tiers?
Yes, and most teams should. Route the volume work to Flash and escalate the hard cases to Pro. It is the same endpoint and the same client — only the model name changes.
Why does Pro not support images?
That is how DeepSeek built the tiers. Practically it means a vision-dependent workload runs on Flash, which is also the cheaper tier — so the constraint rarely bites.
Is 1M context real, or marketing?
It is the published figure for both tiers, and it is why you can send a whole contract or claims file in one call instead of building a chunking pipeline. Test it on your longest document before you rely on it.
Do you serve any other models?
No. We serve the DeepSeek V4 family. If you need a broad catalog behind one key, a router like OpenRouter is the better fit — we wrote up that comparison too.
TWO TIERS, ONE ENDPOINT
Input
Your workload, by shape
high volume or hard reasoning
DeepSeek V4.1 Flash / V4.1 Pro
Output
Flash or Pro, same endpoint
swap the model name only
- Flash: image and video input
- Pro: reasoning, no vision
- Both: 1M context, 384K output
Tell us what you're running.
Send us your monthly volume and what you use today. We usually reply within one business day with a price and a named contact.