DeepSeek API: Pricing, Keys & How to Call It

DeepSeek runs the cheapest frontier-class API in 2026 — V4-Pro at $0.435 input / $0.87 output per million tokens, a fraction of what OpenAI or Anthropic charge for comparable coding quality. This guide gives you the full, verified price table (including the cache-hit rate most articles miss), how to get a key, the rate limits, and how to call it in one line — directly or through a gateway. Every number is checked against DeepSeek's own pricing page.

What the DeepSeek API is

The DeepSeek API gives you programmatic access to DeepSeek's models — the V4 generation (V4-Pro and V4-Flash) plus the prior V3.2 — over an OpenAI-compatible HTTP endpoint. Its headline draw is price: DeepSeek is the cheapest frontier-class API in 2026, and the models are open-weights (MIT), so you can also self-host. For most teams, though, the hosted API is the practical path, and it's remarkably cheap.

DeepSeek API pricing

Here is the complete, current price table — including the cache-hit input rate that most "DeepSeek API pricing" articles omit, and which is the single biggest cost lever for repeated-context workloads:

Per 1M tokens, USD. From DeepSeek's API pricing page, June 2026. Cache-hit input applies to repeated prompt prefixes (system prompts, long context) and is roughly 100× cheaper than cache-miss.

How it compares on price

The reason DeepSeek's API gets so much search traffic is simple: on output price — the expensive side of any bill — it undercuts the Western frontier by an order of magnitude, for coding quality within a few points. Here's V4-Pro's output price against the other major APIs:

For the quality side of that trade — where DeepSeek sits on independent benchmarks versus Claude and GPT — see the DeepSeek V4 vs GPT-5.5 vs Opus 4.8 comparison.

How to get a DeepSeek API key

1. Sign up at platform.deepseek.com and verify your account.
2. Open the API keys section and create a key. Copy it immediately — it's shown once.
3. Add billing/credit; new accounts often get trial credit to start.
4. Set the key as an environment variable (never hard-code it): export DEEPSEEK_API_KEY=...

That gives you direct access. If you also want to reach Claude, GPT, Gemini and 300+ other models with the same key, a gateway (covered below) is the alternative — one signup instead of one per provider.

How to call the DeepSeek API (one line)

Because the API is OpenAI-compatible, you use the standard OpenAI SDK and just point it at DeepSeek's base URL:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseek.com/v1",
    api_key="$DEEPSEEK_API_KEY",
)

resp = client.chat.completions.create(
    model="deepseek-v4-pro",            # or deepseek-v4-flash for the cheap tier
    messages=[{"role": "user", "content": "Review this function for bugs..."}],
)
print(resp.choices[0].message.content)

The same OpenAI-compatible shape means moving between DeepSeek, a gateway, or self-hosted weights is a base-URL and model-id change — no rewrite.

Is the DeepSeek API free? Rate limits

DeepSeek's own API is paid but very cheap, with trial credit for new accounts. For genuinely free usage, some third-party hosts and OpenRouter expose free, rate-limited endpoints of certain DeepSeek models (for example a free R1 endpoint) — fine for prototyping, not for production throughput. On rate limits, DeepSeek doesn't publish one fixed number; throughput scales with account standing and can tighten during peak demand. If steady production throughput matters, calling DeepSeek through a gateway that pools capacity and can fail over to another model is the more reliable pattern.

Which DeepSeek model to use

Calling DeepSeek through a gateway

If you use more than one provider — or want DeepSeek's price with automatic failover when it's rate-limited — a gateway is the cleaner option. DataLLM Lab carries DeepSeek V3.2 today and reaches 300+ other models through one OpenAI-compatible key, so you can route the cheap bulk to DeepSeek and escalate hard cases to a frontier model without a second integration:

client = OpenAI(base_url="https://www.datallmlab.com/v1", api_key="$DATALLMLAB_API_KEY")
# same code reaches deepseek/deepseek-v3.2, anthropic/claude-opus-4.7, openai/gpt-5.4, ...

FAQ