DeepSeek API pricing

Why the rate is below list

Official DeepSeek pricing is set for direct retail customers. TokModel buys inference capacity at enterprise volume and passes a share of that discount through, which is why the same model costs roughly half here. This is capacity resale, not a re-hosted or quantised variant — the model that answers your request is the model DeepSeek published.

There is no subscription layer on top. You pay per token, exactly as the table above states, and nothing is deducted between calls.

Flash or Pro?

For chat, summarisation, extraction and classification, evaluate a Flash model using your own representative examples. Compare the current input and output rates in the table above.

For multi-step reasoning, refactoring or long-document analysis, compare Pro and Flash on task quality, supported parameters and token usage before choosing. Pricing differences do not establish a quality benchmark.

Switching from the official API

Update the base URL and API key, then select an exact model ID from TokModel. Chat Completions uses an OpenAI-compatible structure; validate model-specific options, tool calls and streaming behavior before migrating. Follow the Python integration and migration guide.

# Python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.tokmodel.com/v1",
    api_key="tm_your_key_here",
)

r = client.chat.completions.create(
    model="deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Summarise this changelog"}],
)
print(r.choices[0].message.content)

# curl
curl https://api.tokmodel.com/v1/chat/completions \
  -H "Authorization: Bearer $TOKMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek-v4.1-flash","messages":[{"role":"user","content":"Hello"}]}'

Keys are created in the console and can be revoked at any time. Full parameter reference, streaming details and the error-code table are in the documentation.

TokModel upstream probe snapshot

TokModel runs an upstream gateway probe every five minutes with a one-token, non-streaming chat request. Success requires HTTP 200 and non-empty choices. Availability is successful probes divided by all probes in the last 24 hours; fewer than 12 samples is shown as unavailable. Latency is the median successful probe time to response headers, not full output generation time. These probes bypass the platform /v1 route and do not measure answer quality or end-to-end customer latency. Source: TokModel pricing and health API. Last probe: 2026-11-12T08:29:42.002Z.

Availability for all models is published on the model pricing page; the page is regenerated from the same API payload every 30 minutes.

Frequently asked