DeepSeek API Pricing Calculator & Complete Cost Guide
Calculate DeepSeek V4 (Flash & Pro) API costs per token and per month. Off-peak pricing is half of peak and covers most hours, so scheduled work runs far below GPT and Claude.
Pricing TLDR
- • 5 million free tokens for new users (no credit card required)
- • Off-peak (most hours) is half of peak: V4.1 Flash $0.15 in / $0.60 out, V4 Pro $0.66 / $1.98 per 1M tokens
- • Peak = 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays; schedule batch jobs off-peak to pay half
DeepSeek API Cost Calculator: Off-Peak vs Peak
Calculate by
Input Tokens
Output Tokens
API Calls / Month
Quick Examples:
Context Caching:
Cache Hit Rate:
50%
(off-peak V4.1 Flash: $0.003 cache hit vs $0.15 miss; V4 Pro: $0.022 vs $0.66)
DeepSeek automatically caches context. When requests share the same prefix, cached segments are reused.
Off-peak vs peak: off-peak rates are half of peak, and off-peak covers most of the week. Peak is only 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays; every other hour, and all weekend, is off-peak. The table shows both, so schedulable work can be planned around the cheaper window.
DeepSeek V4.1 Flash (deepseek-flash)
Context
384K output
Quality
Off-Peak
$0.38/mo
In $0.08 · Out $0.60 /1M
Peak
$0.75/mo
In $0.15 · Out $1.20 /1M
DeepSeek V4 Pro (deepseek-v4-pro)
Context
384K output
Quality
Off-Peak
$1.33/mo
In $0.34 · Out $1.98 /1M
Peak
$2.66/mo
In $0.68 · Out $3.96 /1M
Burning through DeepSeek API credits?
Track your DeepSeek API spending in real-time.
Free 7-day trial. No sign-up, no credit card.
About DeepSeek API
What is DeepSeek API?
The DeepSeek API provides programmatic access to DeepSeek's V4 family: V4.1 Flash (cost-effective, with image input, and currently the top-quality model) and V4 Pro (the larger model). V4 bills on a two-tier peak/off-peak schedule where off-peak is half price and covers most hours, with a 1M-token context window, 384K max output, and automatic context caching. Thinking mode is a per-call setting on any model, not a separate price. New integrations should use the `deepseek-flash` and `deepseek-v4-pro` model names rather than the older `deepseek-chat` and `deepseek-reasoner` aliases or the retired `deepseek-v4-flash` names.
- Off-Peak Pricing at Half Rate: DeepSeek charges half rate outside peak hours (peak is 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays). Off-peak covers most of the week, so schedulable work runs at V4.1 Flash $0.15/$0.60 or V4 Pro $0.66/$1.98 per 1M tokens, half the peak rate. Batch and background jobs are the natural fit.
- Extremely Cost-Effective: Even at peak, V4.1 Flash is a fraction of frontier pricing; off-peak it runs around 97% cheaper than GPT-5.5 ($5/$30). V4 Pro off-peak ($0.66/$1.98) also undercuts Claude Sonnet.
- Automatic Context Caching: Context caching is on by default across all tiers. When requests share a prefix with recent ones, cached segments are retrieved automatically. Off-peak, cache hits cost $0.003/M on V4.1 Flash (versus $0.15 for a miss) and $0.022/M on V4 Pro.
When to Use DeepSeek API
DeepSeek is ideal for cost-sensitive applications that need good capability without frontier pricing. Use V4.1 Flash for most work, including high-volume production and image input, and try V4 Pro only where testing shows it does better on your task. Where the work is schedulable, run it in off-peak windows to pay half.
Ideal for
- Cost-sensitive production applications
- High-volume batch and background jobs (schedule off-peak for half price)
- Math, logic, and coding tasks (turn on thinking mode)
- General chatbots and content generation
- Applications with repetitive prompts (benefits from caching)
Not ideal for
- Applications requiring maximum frontier capability
- Use cases needing specific tool integrations not offered
- Latency-critical work locked to peak hours (no off-peak discount)
- Workloads requiring strict SLAs or guaranteed uptime
DeepSeek API Pricing Breakdown
Free Tier
New users receive 5 million free tokens upon registration with no credit card required. These credits are automatically applied to your usage and work across all models.
- Sign up at platform.deepseek.com, no credit card required
- Receive 5 million free tokens instantly
- Credits work across all models (V4.1 Flash and V4 Pro)
- Additional credits can be purchased as needed
Key Features
Peak / Off-Peak Pricing
DeepSeek V4 bills at two rates: off-peak (half price, most hours) and peak (01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays). Off-peak per 1M tokens is V4.1 Flash $0.003/$0.15/$0.60 and V4 Pro $0.022/$0.66/$1.98; peak is exactly double. Moving batch and background work into off-peak windows halves the bill, which is the biggest lever on DeepSeek cost.
Automatic Context Caching
All requests benefit from context caching automatically. When prompts share the same prefix, cached content is reused. Off-peak, V4.1 Flash cache hits cost $0.003/M versus $0.15/M for a miss, roughly a 98% saving on cached input.
1M Context, 384K Max Output
V4 ships with a 1M-token context window and 384K max output across every model and mode, a major expansion from V3.2's 128K context and split 8K-64K outputs.
OpenAI- and Anthropic-Compatible API
DeepSeek API speaks two formats: OpenAI-compatible at the default base URL and Anthropic-compatible at `https://api.deepseek.com/anthropic`. Migrate from either ecosystem by changing your base URL and API key.
DeepSeek API Monthly Cost Estimates
Light Use
$1-8/mo
• Personal projects
• <1K requests/day
• V4.1 Flash off-peak
Medium Use
$8-40/mo
• Small apps
• 1-5K requests/day
• Off-peak where possible
Heavy Use
$40-200/mo
• Production apps
• 5-20K requests/day
• Schedule batch off-peak
Enterprise
$200+/mo
• Large scale
• 20K+ requests/day
• High cache + off-peak
7 DeepSeek API Cost Optimization Tips
Schedule Batch Jobs for Off-Peak
Off-peak rates are half of peak, and off-peak covers most of the week (peak is only 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays). Move batch, background, and non-urgent jobs into off-peak windows to pay half. This is the single biggest lever on a DeepSeek bill.
Maximize Cache Hit Rate
Structure prompts with consistent prefixes (system prompts, instructions) to maximize cache hits. Off-peak on V4.1 Flash, cache hits cost $0.003/M versus $0.15/M for misses. Aim for 70%+ cache hit rates in production.
Start with V4.1 Flash, Test Pro Selectively
V4.1 Flash handles classification, summarization, general chat, and most coding at roughly 3-4x lower cost than V4 Pro, and it currently scores higher on quality. Move a workload to V4 Pro only after testing shows it does better on that task.
Turn Thinking Mode On Only When It Pays
Thinking mode is a per-call setting, not a separate price, but it spends more output tokens on hidden reasoning. Enable it for hard math, logic, and code where Chain-of-Thought lifts quality; leave it off for classification, summarization, and general queries.
Optimize Output Length
V4 supports up to 384K output tokens, but cost accumulates per token. Set appropriate max_tokens limits. Most tasks finish well under 10K output tokens.
Compare with Competitors
V4.1 Flash off-peak at $0.15/$0.60 is around 95-98% cheaper than GPT-5.5 ($5/$30) and Claude Sonnet ($3/$15). V4 Pro off-peak at $0.66/$1.98 also comes in below Claude Sonnet.
Monitor Token Usage
Track your cache hit rates and token consumption via the DeepSeek platform. Understanding your caching patterns and peak-vs-off-peak split helps optimize prompt design and reduce costs further.
DeepSeek Model Selection Guide
Use Case
Customer Support Chat
Recommended Model
V4.1 Flash
Cost-effective tier
Monthly Cost (Est.)
~$2-8
Why This Model?
Fast and cheapest for general queries
Use Case
Code Generation
Recommended Model
V4.1 Flash
Thinking mode on
Monthly Cost (Est.)
~$5-20
Why This Model?
DeepSeek's top-quality model, with Chain-of-Thought at Flash pricing
Use Case
Hard Math & Logic
Recommended Model
V4.1 Flash
Thinking mode on
Monthly Cost (Est.)
~$5-20
Why This Model?
Top quality score in the lineup; test V4 Pro only if Flash falls short
Use Case
Image Understanding
Recommended Model
V4.1 Flash
Image input
Monthly Cost (Est.)
~$3-15
Why This Model?
Accepts images at Flash pricing (billed as input tokens)
Use Case
Data Extraction
Recommended Model
V4.1 Flash
Cost-effective tier
Monthly Cost (Est.)
~$2-8
Why This Model?
JSON output, tool calling, lowest cost
Use Case
High-Volume Batch
Recommended Model
V4.1 Flash
Off-peak + high cache
Monthly Cost (Est.)
~$4-40
Why This Model?
Schedule off-peak and maximize cache hits for the lowest cost
Start Tracking Your DeepSeek API Spending
Monitor your DeepSeek V4 credit balance and per-tier usage from your menubar. Alerts before you run out.
Free 7-day trial. No sign-up, no credit card.
DeepSeek API Pricing FAQ
Common questions about DeepSeek API costs, billing, and optimization
DeepSeek V4 bills at two rates, off-peak and peak, where off-peak is half of peak. Off-peak per 1M tokens: V4.1 Flash is $0.003 cache hit / $0.15 cache miss / $0.60 output, and V4 Pro is $0.022 / $0.66 / $1.98. Peak is double those. Both share a 1M-token context window and 384K max output, with automatic context caching on by default.
Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays. Every other hour, plus the whole weekend, is off-peak at half the peak rate. Since off-peak covers most of the week, moving batch and background work into off-peak windows is the single biggest lever on a DeepSeek bill.
V4.1 Flash (model name `deepseek-flash`) is the cost-effective tier for high-volume production, chat, summarization, and most coding, and it accepts image input. It is also DeepSeek's top-quality model right now. V4 Pro is the larger model at roughly 3-4x the per-token cost, and it now trails V4.1 Flash on quality scores, so test it against Flash before paying the premium. The older V4 Flash and V4 Flash Vision are retired; their model names now route to V4.1 Flash. Both current models share a 1M context window and 384K max output, and thinking mode is a per-call setting on either one, not a separate price.
Far cheaper. V4.1 Flash off-peak ($0.15 input / $0.60 output per 1M) is around 95-98% below GPT-5.5 ($5/$30) and Claude Sonnet ($3/$15), and even at peak ($0.30/$1.20) it stays a fraction of the cost. V4 Pro off-peak ($0.66/$1.98) also undercuts Claude Sonnet. DeepSeek gets there with a Mixture-of-Experts architecture, Sparse Attention, and Multi-head Latent Attention.
On V4.1 Flash off-peak, light personal use runs a few dollars a month, medium production apps $10-60/month, and heavy workloads $60-300/month. V4 Pro costs roughly 3-4x more per token, and peak hours double whichever tier you use. Use the calculator above to see your off-peak and peak monthly cost side by side for your own volume.
DeepSeek gives new users 5 million free tokens on signup, no credit card required. The credits work across all models (V4.1 Flash and V4 Pro). Buy more on the platform via credit card or PayPal when you need them.
