DeepSeek V3.2 API pricing explained: what a million tokens costs

What does "per million tokens" actually mean?

When you see DeepSeek V3.2 API pricing listed as "$0.40 per million input tokens" and "$0.80 per million output tokens", it helps to translate that into real-world numbers.

A token is roughly 0.75 English words. So one million tokens is about 750,000 words — roughly three full-length novels. Most production API calls use far less. A typical chatbot response might be 200–500 tokens, and a document summarization task might consume 2,000–8,000 input tokens.

Here's what common usage scenarios cost with DeepSeek V3.2 at TokenShop's pricing:

Scenario	Input tokens	Output tokens	Input cost	Output cost	Total
Short Q&A	200	150	$0.00008	$0.00012	$0.00020
Code review	1,500	500	$0.00060	$0.00040	$0.00100
Document analysis	8,000	1,000	$0.00320	$0.00080	$0.00400
Batch processing (1k docs)	8M	1M	$3.20	$0.80	$4.00

The key takeaway: for interactive use, individual requests cost fractions of a cent. Costs become noticeable only at batch scale.

Input vs output pricing: why the difference?

DeepSeek V3.2 charges $0.40 per million input tokens and $0.80 per million output tokens — output is exactly double the input rate. This is common across open-weight models and reflects the asymmetric compute cost.

Generating tokens requires the model to run autoregressively: each new token depends on all previous tokens, so output generation is inherently more expensive than processing a fixed input. The 2x ratio is actually modest compared to some competing models where output can cost 3–8x input.

For context, here's how DeepSeek V3.2 compares to other models available on TokenShop:

• Qwen3 32B: input $0.30, output $0.90 (3x ratio, cheaper input but pricier output)
• GLM-4.6: input $0.80, output $2.40 (3x ratio, higher floor but 202K context)

DeepSeek V3.2 sits in the middle — cheaper on output than Qwen3 32B, but slightly more expensive on input. The 2x ratio makes it more predictable for applications that generate longer responses.

How to estimate your monthly bill

To calculate your approximate monthly cost, track three numbers:

1. Average input tokens per request (prompt + system message + conversation history)
2. Average output tokens per response (generated text)
3. Number of API calls per month

Formula:

(input_tokens × $0.40 + output_tokens × $0.80) × (calls / 1,000,000)

For a customer support bot handling 50,000 conversations per month, with 1,500 input tokens and 300 output tokens per conversation:

(1500 × $0.40 + 300 × $0.80) × (50000 / 1000000)
= (600 + 240) × 0.05
= 840 × 0.05
= $42.00 per month

That's about $0.00084 per conversation — well under a tenth of a cent.

Token counting in practice

The most common surprise for new users is that token counts aren't the same as character or word counts. You can verify this yourself with a quick API call using the tokenizer endpoint:

curl https://tokshop.xyz/v1/tokenize \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v3.2",
    "text": "Explain the difference between supervised and unsupervised learning in one paragraph."
  }'

The response will return a token count and the individual token IDs. For most OpenAI-compatible SDKs, you can also use the tiktoken library with the appropriate encoding to estimate tokens client-side before sending requests.

Context window and cost implications

DeepSeek V3.2 supports a 131,072-token context window. This is generous — enough for roughly 100 pages of text. However, longer contexts increase input cost linearly. A single request using 100K input tokens costs:

100,000 × $0.40 / 1,000,000 = $0.04

That's still only four cents per request, but it adds up if you're doing many long-context operations. For comparison, GLM-4.6 offers 202,752 tokens of context at $0.80/M input — so DeepSeek V3.2 is half the per-token input cost for a still-large context.

A practical strategy: use the full context window for retrieval-augmented generation (RAG) where you need to include many document chunks, but avoid sending redundant conversation history if you're using short-turn chat.

Getting started with DeepSeek V3.2 on TokenShop

TokenShop provides an OpenAI-compatible API, so you can use any OpenAI SDK with a simple endpoint change. The base URL is https://tokshop.xyz/v1.

from openai import OpenAI

client = OpenAI(
    base_url="https://tokshop.xyz/v1",
    api_key="YOUR_API_KEY"
)

response = client.chat.completions.create(
    model="deepseek/deepseek-v3.2",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Write a Python function to merge two sorted lists."}
    ],
    max_tokens=500
)

print(response.choices[0].message.content)
# Check token usage
print(f"Input tokens: {response.usage.prompt_tokens}")
print(f"Output tokens: {response.usage.completion_tokens}")

New users receive a $0.50 trial credit after registering with an email address. The billing system runs in test mode (no real charges), and uses a prepaid micro-USD ledger — you top up, then spend down. When your balance drops below $0.001, the API returns HTTP 402.

For the latest pricing across all available models, including DeepSeek V3.2, visit the TokenShop pricing page.

Conclusion

DeepSeek V3.2's pricing at $0.40/M input and $0.80/M output is competitive for a 131K-context model, especially given the 2x output ratio. For most interactive applications, individual requests cost well under a cent. The real cost analysis matters at scale — batch processing, long-context RAG, or high-traffic chatbots. By tracking your average token usage per request and estimating monthly call volume, you can predict costs accurately before committing to a production deployment.