Deepseek Api Pricing: Complete 2026 Cost Guide

The DeepSeek API bills per token, and every rate doubles during UTC peak hours. Off-peak input starts at $0.15 per 1M tokens on DeepSeek V4.1 Flash and climbs to $3.96 per 1M output tokens on DeepSeek V4 Pro at peak. The flagship V4 Pro costs $0.66 per 1M input tokens and $1.98 per 1M output tokens off-peak.

Our team at Cabina.AI tracks pricing for DeepSeek directly, since we give developers access to DeepSeek V4 Pro alongside 55+ other models through one subscription – when a rate changes, we see it firsthand. We noticed the peak window runs on UTC clock time alone: a call placed at 02:00 UTC bills at the off-peak rate no matter where the developer is located, which is easy to miss if a job gets scheduled by local time instead.

DeepSeek API Pricing Table – All Models

DeepSeek is both the model family and the Hangzhou-based company that sets and bills these prices – the same name covers the API, the open weights, and the free consumer app. DeepSeek currently bills two models through the API: DeepSeek V4.1 Flash and DeepSeek V4 Pro, both with a 1M-token context window and a 384K-token output cap. Every deepseek api pricing figure below is per 1M tokens, split by whether the request lands in a cache hit or a cache miss, and by whether it lands inside DeepSeek peak window.

Peak hours run 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday; every other hour bills at the off-peak rate. Two older names, deepseek-v4-flash and deepseek-v4-flash-vision-exp, still work in existing code, but DeepSeek now routes both to V4.1 Flash and bills them at the Flash price above. DeepSeek V4 Pro was originally scheduled to sunset on September 14, 2026; DeepSeek has since confirmed it will keep serving V4 Pro past that date at the same billing method, so the row above stays live rather than pointing at a retired model.

DeepSeek API Cost Calculator

Pick a model, enter how many input and output tokens the application sends in a typical month, and check the peak-hours or caching boxes that match how the traffic actually runs, to get an estimated deepseek api cost for that month. The widget below pulls its per-token rates straight from the table above, so the two never drift apart.

DeepSeek API Request Example

DeepSeek endpoint speaks the same request format as the OpenAI chat completions API, so an existing OpenAI SDK setup usually only needs a new base URL and API key. Here is a minimal authenticated call to deepseek-flash, the model name to use for DeepSeek V4.1 Flash:

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-api-key-here" \
  -d '{
        "model": "deepseek-flash",
        "messages": [
          {"role": "user", "content": "Summarize this in one sentence."}
        ],
        "stream": false
      }'

DeepSeek also exposes an Anthropic-compatible endpoint at https://api.deepseek.com/anthropic, so code already written against the Claude Messages API can point at DeepSeek by changing the base URL and model name rather than rewriting the request shape. Both models default to a thinking mode that can be turned off per request, and both support JSON output and tool calls; fill-in-middle completion works only outside thinking mode.

DeepSeek API Price and Model-ID History

The two rows in the table above are the only models the DeepSeek API bills today, and that lineup is newer than it looks. DeepSeek retired the deepseek-chat and deepseek-reasoner aliases, the endpoints behind the earlier V3 and R1 generation, on July 24, 2026. deepseek-v4-flash followed on September 10, 2026, replaced in the billing system by V4.1 Flash. A handful of third-party pricing trackers still list deepseek-chat, R1, or V4 Flash 0731 at their old rates – those model IDs no longer bill on the current DeepSeek API, so a quote built on them will not match a live invoice.

Pricing Tiers for DeepSeek, Explained

Off-Peak vs Peak Pricing

Across both deepseek models, api pricing beyond the base rate comes down to one real tier: a time-of-day multiplier, not a named plan. Every price in the table above exactly doubles once a request lands inside the peak window, and drops back to the off-peak number the moment it doesn’t. DeepSeek does not publish a separate Batch API discount the way some other providers do; shifting non-urgent jobs into off-peak hours is the closest equivalent, and it saves the same 50% a batch tier would. Rate limits move with the model, not the clock: DeepSeek V4.1 Flash allows up to 2,500 concurrent requests, while DeepSeek V4 Pro caps out at 500, a fact worth checking before assuming a Flash-sized deployment will run unmodified against Pro.

Prompt Caching on the DeepSeek API

Cache-hit input pricing is the single biggest discount DeepSeek publishes: $0.003 per 1M tokens off-peak on Flash and $0.022 on Pro, both exactly 2% of the matching cache-miss rate. That is a 98% discount on any input DeepSeek has already processed and cached, and it applies identically across both models. From running the DeepSeek API ourselves, this is the number third-party price trackers most often leave out of their headline figures, since most of them quote only the cache-miss rate as “the” input price. A tokenizer note worth pairing with this: DeepSeek uses its own byte-pair tokenizer, so how much is a token in DeepSeek is not the same question as how much is a token in ChatGPT or Claude – identical text can produce a different token count on each, which changes the real bill even when the per-token rate looks similar on paper.

DeepSeek V4 Pro vs DeepSeek V4.1 Flash: Which Should You Use?

Flash is the cheaper default, and the only one of the two that accepts image input alongside text. Pro costs roughly 4.4x as much per input token off-peak (3.3x on output) and drops vision support entirely, so it earns its price on harder reasoning and agentic work rather than on everyday chat or document Q&A. In our view, the deepseek api costs per model line up cleanly with that split: start every new integration on Flash, and move a specific workload to Pro only after Flash has measurably fallen short on it, rather than defaulting to the pricier model up front.

DeepSeek V4 Pro and V4.1 Flash also run on Microsoft Foundry alongside the first-party API, giving teams already committed to Azure a second place to bill the same models. If a workload needs to run there for compliance or procurement reasons, the per-token math above still applies; only the invoice changes hands.

Is the DeepSeek API Free?

Not for production use – every request is billed per token from the first call, with no ongoing free quota. Multiple third-party trackers report that new accounts receive a one-time credit of 5 million tokens, valid for 30 days, but this figure is not stated on DeepSeek own pricing or quickstart documentation, so treat it as unconfirmed until checked directly in the DeepSeek platform console before publishing an exact number. Once that credit runs out, how much do deepseek tokens cost depends entirely on whether the traffic lands off-peak or inside one of the two daily peak windows.

If you’d rather not commit to a single provider while testing, DeepSeek V4 Pro is also available on Cabina.AI alongside ChatGPT, Claude, and 50+ other models under one subscription – useful for comparing outputs before committing to an API integration.

DeepSeek API vs ChatGPT API Pricing

Set deepseek api pricing per million tokens next to OpenAI own table and DeepSeek reputation as the cheapest frontier-grade API holds at the flagship tier, not at the very bottom of either lineup: OpenAI smallest model still undercuts DeepSeek smallest model on paper.

Line up the flagships instead of the entry-level models and the gap flips hard: gpt-6-astra costs roughly 7-13x more per token than DeepSeek V4 Pro at the off-peak rate, and the difference widens further against OpenAI older o1-pro tier, which lists at $75 input and $300 output per 1M tokens. From tracking both APIs side by side, that flagship-to-flagship gap is where DeepSeek price advantage actually shows up, not at the nano/mini end of either lineup.

DeepSeek API Pricing FAQ

Is the DeepSeek API free?

No. Every request is billed per token from the first call. New accounts may receive a one-time signup credit, but DeepSeek does not publish an ongoing free tier for production use.

How much does the DeepSeek API cost per 1M tokens?

Off-peak, DeepSeek V4.1 Flash costs $0.15 per 1M input tokens and $0.60 per 1M output tokens; DeepSeek V4 Pro costs $0.66 input and $1.98 output. Every rate doubles during UTC peak hours, and cached input drops to 2% of the standard rate.

What’s the cheapest and most expensive DeepSeek model?

DeepSeek V4.1 Flash is the cheapest at $0.15 per 1M input tokens off-peak. DeepSeek V4 Pro is the most expensive, at $3.96 per 1M output tokens during peak hours.

Does DeepSeek offer a caching or off-peak discount?

Yes to both. Cached input bills at 2% of the standard rate on either model, a 98% discount, and every price is automatically halved outside DeepSeek UTC peak window. DeepSeek does not run a separate Batch API discount on top of these.

What is the newest DeepSeek model and how much does it cost?

DeepSeek V4 Pro is the current flagship, priced at $0.66 per 1M input tokens and $1.98 per 1M output tokens off-peak. DeepSeek V4.1 Flash is the newest release in the cheaper tier, replacing the retired V4 Flash at the same price point.

Why is DeepSeek API pricing cheaper during off-peak hours?

DeepSeek discounts off-peak traffic to smooth demand on its own infrastructure. The window is fixed by UTC clock time, 01:00-04:00 and 06:00-10:00 UTC on weekdays counting as peak, so a deployment can cut its bill in half just by shifting non-urgent jobs outside those hours.