deepseek-v4-pro API — per-token pricing and code examples

Checking statusPrice version 4 · in effect since Nov 19, 2026

Price per 1M tokens

InputOfficial price $0.66$0.165
Cache readOfficial price $0.022$0.0055
Cache write$0.165
OutputOfficial price $1.98$0.495

Minimum per request$0.00002

Cache rates apply when the provider reports cache tokens.

75% off

About this model

deepseek-v4-pro is served through OpenAI Chat Completions, OpenAI Responses, Anthropic Messages. You pay $0.165 per 1M input tokens and $0.495 per 1M output tokens. That is 75% below the official price of $0.66 and $1.98. Cache reads cost $0.0055 per 1M tokens when the provider reports cached tokens. It is included in budget packs from $9.99 up.

Where DeepSeek-V4-Pro fits

DeepSeek-V4-Pro is one of the two models DeepSeek launched with its V4 generation. DeepSeek first released it on April 24, 2026 and announced general availability on August 13, 2026 across its app, web and API. The version DeepSeek documents today is DeepSeek-V4-Pro-0813, and its update note for that release says agent capabilities were improved. DeepSeek later stated that V4-Pro API service continues past September 14, 2026.

The limits are a 1M-token context and up to 384K tokens of output. Input is text only: DeepSeek's feature table marks vision as not supported for V4-Pro, so screenshots and scanned pages need a different model, for example DeepSeek-V4.1-Flash, which does accept images. With text-only input and the agent improvements DeepSeek describes, V4-Pro suits text-heavy agent work such as long code reviews, planning a refactor across many files or working through a dense specification.

Thinking modes and the top_p trap

DeepSeek-V4-Pro thinks by default, at high effort. The August 13 update introduced three effort levels for it: low, high and max. You can also turn thinking off. On /v1/chat/completions the switch is thinking: {"type": "enabled"} or {"type": "disabled"}, and the depth is reasoning_effort. DeepSeek maps other effort names onto its three levels, for example medium to high and xhigh to high.

The parameter that surprises people is top_p. DeepSeek's thinking guide says that for deepseek-v4-pro, your top_p is ignored in thinking mode: it is fixed at 0.95 when tools are present and at 1.0 otherwise. Only in non-thinking mode does your value take effect. If you tune sampling for creative output, run V4-Pro with thinking disabled. In thinking mode, temperature, presence_penalty and frequency_penalty are also accepted without error but change nothing.

Tool loops have one more rule. When the request includes tools, DeepSeek asks you to send back the reasoning_content from every previous turn. Agent frameworks that strip reasoning from history need to keep it for V4-Pro.

{
  "model": "deepseek-v4-pro",
  "reasoning_effort": "max",
  "messages": [{ "role": "user", "content": "Plan the migration of this module to the new storage layer" }]
}

The body above is for /v1/chat/completions. On DeepSeek's own Responses API, the same thing is controlled with reasoning.effort, where none turns thinking off.

What it costs on Cent AI

Billing is per token from your prepaid USD balance, using the four rates in the price box on this page. In thinking mode DeepSeek-V4-Pro produces reasoning tokens before its answer, and Cent AI bills them as output. Because thinking is on unless you disable it, a plain request already includes that reasoning cost. Use max effort for the hard problems and low, or thinking off, for routine prompts.

DeepSeek's context cache is automatic and best-effort: on DeepSeek's own API, hits are reported as prompt_cache_hit_tokens, and DeepSeek says cached data is usually cleared within a few hours to a few days. Cent AI applies cache rates only to hits the model reports, never fills them in on its own, and returns the cached tokens in the usage of the endpoint you call: cache_read_input_tokens on /v1/messages, input_tokens_details.cached_tokens on /v1/responses.

Using it from your tools

DeepSeek-V4-Pro works on /v1/chat/completions, /v1/responses and /v1/messages. Cent AI does not change your parameters silently. On DeepSeek's own Anthropic-compatible API, DeepSeek documents thinking and output_config.effort, and lists budget_tokens, top_k and cache_control as ignored.

Specification

Context window
—
Input and output
Text
Endpoints
/v1/chat/completions, /v1/responses, /v1/messages
Smallest pack
$9.99

Latency and error rate per model are on the status page. status.cent-ai.dev

What a request costs

Open the full calculator

Cost per request

90% is an assumption you can change, not a promise. Cache is billed only from cache tokens the model provider reports.

See how many such requests it covers.

Estimate

deepseek-v4-pro
Fresh input99,000 tok
Cache read891,000 tok
Output10,000 tok

Per request$0.0261
At official pricesAt official prices $0.1047

Total$0.0261

An estimate at price version 4. Your bill comes from the usage the provider reports for each request.

Save $0.0785

Code

curl https://api.cent-ai.dev/v1/chat/completions \
  -H "Authorization: Bearer $CENT_AI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-v4-pro", "messages": [{"role": "user", "content": "Hello"}]}'

Set CENT_AI_KEY to a key from the dashboard. OpenAI SDKs use https://api.cent-ai.dev/v1; Anthropic SDKs use https://api.cent-ai.dev.

Price history

Version 4, in effect since Nov 19, 2026.

  1. Effective

    per 1M tokens

    Inputold $0.066 new $0.165
    Cache readold $0.0066 new $0.0055
    Cache writeold $0.066 new $0.165
    Outputold $0.198 new $0.495
  2. Effective

    per 1M tokens

    Inputold $0.053 new $0.066
    Cache readold $0.0053 new $0.0066
    Cache writeold $0.053 new $0.066
    Outputold $0.158 new $0.198
  3. Effective

    per 1M tokens

    Inputold $0.053 new $0.053
    Cache readold $0.001766667 new $0.0053
    Cache writeold $0.053 new $0.053
    Outputold $0.158 new $0.158

Compare with

All models against official prices

Questions

How much does deepseek-v4-pro cost on Cent AI?

$0.165 per 1M input tokens and $0.495 per 1M output tokens. Cache reads cost $0.0055 and cache writes $0.165 per 1M tokens. Each request is rounded up once.

Is there a minimum charge?

A successful request costs at least $0.00002. A request that ends early, because the client left or the output was cut off, pays only for its tokens.

How do I pay?

Top up a prepaid USD balance with QRIS or USDT/USDC, or buy a budget pack once. Nothing renews automatically.

What if the provider does not report usage?

The request is billed from an estimate and marked estimated on the Requests page. An estimate never contains cache tokens.

Which endpoint do I call?

/v1/chat/completions, /v1/responses, /v1/messages. The OpenAI base URL is https://api.cent-ai.dev/v1.

Try deepseek-v4-pro with one key

One prepaid balance for every model. Pay with QRIS or USDT/USDC.

Start free