DeepSeek V4.1 Flash API pricing and cost calculator

By TechCompare · Updated · How we check

A half-cached 100K-input, 5K-output request costs $0.006285 at the cheapest host, Wafer, or $629 for 100,000 calls. Check your host's own rates before you budget.

A newer snapshot of DeepSeek V4.1 Flash is out: DeepSeek V4.1 Flash 1017. Price DeepSeek V4.1 Flash 1017 in the calculator or compare every DeepSeek model.

Calculator

Cost Comparison

Based on 100,000 input tokens (50% cached), 5,000 output tokens, and 100 requests.Prices come live from OpenRouter at each maker's own price where it's listed. Temporary deals are left out.

Loading model prices

OpenRouter lists no DeepSeek-run endpoint for DeepSeek V4.1 Flash, so it's sold by third-party hosts that set their own prices. The chart starts from the cheapest host for this request and names it, so the estimate matches the calculations below. Rates depend on the host and its quantization. The host you route to can change the bill a lot even when the model, prompt, cache usage, and generated output are identical.

DeepSeek: DeepSeek V4.1 Flash

Cheapest host: Wafer (mxfp4)

Input
$0.051
Output
$0.567
Cached
$0.018

per 1M tokens

Batch availableFlex availableOff-peak available

Price checked 2026-11-19

How this is calculated

The cheapest listed host, Wafer with mxfp4 weights, charges $0.051 input, $0.018 cache-hit input and $0.567 output per million tokens. For 100,000 input tokens and 5,000 output tokens without cache hits, that's $0.007935 per request. If half the input hits the cache, the same request costs $0.006285.

The chart's 100 monthly requests therefore cost about $0.63, and 100,000 half-cached requests come to about $629. Price the host you'll actually call, since every host sets its own rates.

Cheaper hosts often serve lower-precision weights such as fp4 or fp8 and differ in their maximum output, from 131K to 943K tokens, so compare the quantization, output limit, and uptime before sending production traffic to the lowest price. Most hosts offer the 1M-token context window. Extra context and generated reasoning still add billable tokens. The chart shows the normal-speed standard tier, so flex, batch, and priority prices are separate.

Verdict

Pick the host before you budget. Start with the cheapest listing for the workload you plan to run, then check that its quantization, output limit, and uptime are acceptable for your task.

Cache reuse reduces the input portion on every host, but hosts differ in how deeply they discount cached input. For comparisons with another API, use the host your application will actually call and count all reasoning output, tool turns, and retries required to get an accepted result.

More API Standalones scenarios

Frequently asked questions

What are DeepSeek V4.1 Flash's API prices?
They depend on the host. The cheapest listing for this page's request, Wafer, charges $0.051 input, $0.018 cache-hit input, and $0.567 output per million tokens, and other hosts set their own rates. Apply the rates of the host you actually call to each request's billed usage.
Why do hosts charge such different prices for the same model?
Hosts differ in hardware, capacity, and the numeric precision of the weights they serve. The lowest prices usually come from fp4 or fp8 quantization, and some hosts also run temporary promotions. Output limits and uptime differ too, so the cheapest listing isn't automatically the right choice for production traffic. Check the quantization shown next to the host name before you rely on its price.
Which API model name selects V4.1 Flash?
On OpenRouter the model ID is deepseek/deepseek-v4.1-flash, and each host accepts requests routed through it. DeepSeek's own API uses different identifiers and later builds of the Flash model, so check the model version returned by the service rather than assuming an ID from one provider is accepted on another.
What is DeepSeek V4.1 Flash 1017?
It's the October 17, 2026 snapshot of DeepSeek V4.1 Flash, which DeepSeek serves under the deepseek-flash API model name. OpenRouter lists it separately as deepseek/deepseek-v4.1-flash-1017, at $0.30 input, $0.006 cache-hit input, and $1.20 output per million tokens. Cheaper hours bring that down to $0.15 input and $0.60 output, so check DeepSeek's pricing page for the current window. These rates are separate from the host listings this page starts from, so settle on one model ID before you budget.
How do I estimate a mix of hosts?
Calculate each host separately and add the totals. For 100,000 requests with 100,000 input tokens, half cached, and 5,000 output tokens, 70,000 calls at Wafer's rates cost about $440. Price the other 30,000 calls at the second host's own rates, then add the two. If the hosts see different token usage, use their own input and output averages.
Does choosing a cheaper host replace prompt caching?
No. The host sets the base rates and caching replaces the uncached-input rate with the cached-input rate for tokens that actually hit the cache. For a fully cached 100,000-token prompt and 5,000 output tokens, the estimate is $0.004635 at Wafer's rates.