DeepSeek API pricing for every model, newest first
By TechCompare · Updated · How we check
DeepSeek V4.1 Flash 1017 is the newest DeepSeek model and the lowest cost one DeepSeek sells for a 100,000 token prompt with a 5,000 token answer, at $0.01395 a call with three quarters of the input cached.
Every DeepSeek model and its API price
Newest first. Prices are per 1M tokens. One request is 100K tokens in and 5K out, with 75% of the input cached.
| Model | Input | Output | Cached | Context | One request | Listed |
|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash 1017DeepSeek | $0.30 | $1.20 | $0.006 | 1.05M | $0.014 | Oct 17, 2026 |
| DeepSeek V4.1 FlashCheapest host: Wafer | $0.051 | $0.567 | $0.018 | 1.05M | $0.0055 | Sep 10, 2026 |
| DeepSeek V4 Flash Vision ExpCheapest host: DeepInfra | $0.44 | $1.32 | $0.014 | 1.05M | $0.0187 | Aug 21, 2026 |
| DeepSeek V4 Pro 0813DeepSeek | $1.32 | $3.96 | $0.044 | 1.05M | $0.0561 | Aug 12, 2026 |
| DeepSeek V4 Flash 0731Cheapest host: Relace | $0.07 | $0.14 | $0.014 | 1.05M | $0.0035 | Jul 31, 2026 |
| DeepSeek V4 Flash 0423Cheapest host: Relace | $0.014 | $0.45 | $0.014 | 1.05M | $0.0037 | Apr 24, 2026 |
| DeepSeek V4 Pro 0423Cheapest host: SiliconFlow | $1 | $1.50 | $0.135 | 1.05M | $0.0426 | Apr 24, 2026 |
| DeepSeek V3.2Cheapest host: Baidu | $0.28 | $0.42 | $0.028 | 164K | $0.0112 | Dec 1, 2025 |
| DeepSeek V3.2 ExpCheapest host: SiliconFlow | $0.27 | $0.41 | $0.27 | 164K | $0.0291 | Sep 29, 2025 |
| DeepSeek V3.1 TerminusCheapest host: SiliconFlow | $0.27 | $1 | $0.27 | 164K | $0.032 | Sep 22, 2025 |
| DeepSeek V3.1Cheapest host: DeepInfra | $0.25 | $0.95 | $0.13 | 164K | $0.0208 | Aug 21, 2025 |
| R1 0528Cheapest host: DeepInfra | $0.50 | $2.15 | $0.35 | 164K | $0.0495 | May 28, 2025 |
Listed is the day OpenRouter added the model. Prices checked Nov 14, 2026.
DeepSeek is an AI lab in Hangzhou, China, and it sells its models through its own API. This page lists all 14 DeepSeek models OpenRouter carries, newest first. A row shows DeepSeek's own price where DeepSeek still sells the model, and the host that charges least where it doesn't, which covers most of the older V3 and R1 releases.
DeepSeek V4.1 Flash 1017 is the fast everyday model, at $0.30 input and $1.20 output per million tokens. DeepSeek V4 Pro 0813 is built for hard reasoning and agent work, at $1.32 and $3.96.
Calculator
Cost Comparison
Based on 100,000 input tokens (75% 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.
How to read the DeepSeek prices
DeepSeek caches on its own. When the start of a prompt matches one it saw recently, the matching tokens bill at the cache hit rate, with no setup. On DeepSeek V4.1 Flash 1017 a hit costs $0.006 per million, 98% off a miss. With three quarters of the example prompt cached, the cached part costs $0.00045 and the rest of the input costs $0.0075.
DeepSeek also has off-peak hours: any time outside 9:00-12:00 and 14:00-18:00 Beijing time (UTC+8). In those hours DeepSeek V4.1 Flash 1017 bills $0.15 input and $0.60 output per million, 50% off input. The peak hours are the Chinese working day, so the cheap window covers the afternoon and evening in Europe and the whole US working day. DeepSeek V4 Pro 0813 has off-peak rates too, at $0.66 input and $1.98 output per million.
DeepSeek's own API calls its Flash model deepseek-flash, and that name moves to each new Flash release. So retest your prompts when one ships, even if your code didn't change.
DeepSeek has released open weights for its main models, which is why so many hosts sell the older ones. DeepSeek V4.1 Flash 1017 reads up to 1.05M tokens, as much as any model in the family.
Verdict
DeepSeek V4.1 Flash 1017 is the newest DeepSeek model and the lowest cost one DeepSeek sells for the example request, so it's the place to start. Keep DeepSeek V4 Pro 0813 for the hard reasoning jobs.
Keep the start of each prompt the same so the cache catches it. And move work that can wait into the off-peak hours.
More LLM pricing scenarios
Related guides
Frequently asked questions
How much does the DeepSeek API cost?
When are DeepSeek's off-peak hours?
How does caching work on the DeepSeek API?
What is deepseek-flash?
Can I run DeepSeek models myself?
Which DeepSeek model has the largest context window?
Where do these prices come from?
Related tools
LLM VRAM Calculator
Calculate the VRAM needed to run or fine-tune any LLM at any quantization.
Use toolPower Cost Estimator
Estimate annual electricity costs for your PC, Server, or TV.
Use toolData Transfer Calculator
Estimate transfer times for files over USB, WiFi, Ethernet, and more.
Use toolJSON Formatter
Validate, format, and minify JSON. Browse it as a tree, run JSONPath queries, compare two documents, or validate against a JSON Schema.
Use tool