DeepSeek V4 Pro: Capabilities, API Access, Pricing, and What to Verify

Short answer: DeepSeek’s documentation describes a V4 Pro API model, but a separate official routing notice says requests using deepseek-v4-pro are routed to V4.1-Flash and billed at Flash rates until V4.1-Pro launches. I found no official announcement of that launch in the sources checked. The API pricing page lists both Pro-0813 and V4.1-Flash-1017 rates, but those listings alone do not establish which model will serve a request to the Pro alias. So, do not treat a successful call to deepseek-v4-pro as proof that you evaluated V4 Pro; confirm the served model and billing with DeepSeek first. (DeepSeek routing notice; DeepSeek API pricing) (deepseek.com)

What DeepSeek V4 Pro is

DeepSeek’s April 2026 preview described V4 Pro as a 1.6-trillion-parameter model with 49 billion active parameters and a 1-million-token context. DeepSeek later announced general availability on August 13, 2026. Its API pricing page identifies the listed API version as DeepSeek-V4-Pro-0813. These are separate preview and availability milestones, not one release date. (DeepSeek V4 preview; DeepSeek V4 Pro GA announcement) (deepseek.com)

The API pricing page lists a maximum output of 384K tokens and says V4 Pro does not support vision. It lists support for thinking and non-thinking modes, JSON output, tool calls, the Responses API, and the Anthropic API format. Treat these as specifications for the listed Pro API version—not confirmation that a request to the deepseek-v4-pro alias is currently served by that version. (DeepSeek API pricing and model details) (api-docs.deepseek.com)

Workload fit: evaluate, don’t assume

DeepSeek introduced V4 Pro with an emphasis on agent capabilities. That is a reason to test it on multistep reasoning, coding, or long-context tasks, not a guarantee of performance on your workload. Compare models using your own prompts, tools, and success criteria. Measure task quality and repair effort, behavior on unanswerable requests, end-to-end latency, tool-call success, and token use—including retries and long outputs.

A DeepInfra article reports a 94% hallucination rate on the named AA-Omniscience benchmark. It does not provide enough methodology to treat that figure as independently validated or representative of every task. It is a benchmark-specific result, not a universal error rate; use it as a reason to test abstention behavior in confidence-sensitive workflows. (DeepInfra’s V4 Pro overview)

V4 Pro versus the current Flash API listing

DeepSeek’s pricing page lists deepseek-v4-pro as Pro-0813 and deepseek-flash as V4.1-Flash-1017. Separately, DeepSeek’s routing notice says that, starting at 04:00 UTC on September 14, 2026, requests using deepseek-v4-pro route to V4.1-Flash at Flash rates until V4.1-Pro launches. The notice is conditional and does not itself establish when that condition ends. I found no official V4.1-Pro launch announcement in the sources checked; the pricing table’s Pro-0813 row does not, on its own, confirm that the alias currently serves Pro-0813. Verify the route for your account before using the alias to evaluate V4 Pro or estimate its cost. (DeepSeek routing notice; DeepSeek API pricing) (deepseek.com)

Production API access

DeepSeek documents deepseek-v4-pro as a model ID for its Chat Completions API, with the endpoint https://api.deepseek.com/chat/completions and bearer-token authentication. This example shows how to call the documented API. It does not establish that the request evaluates V4 Pro: the routing notice says the alias is directed to V4.1-Flash until V4.1-Pro launches. (DeepSeek Chat Completions API; DeepSeek routing notice) (deepseek.com)

export DEEPSEEK_API_KEY="your-api-key"

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${DEEPSEEK_API_KEY}" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [
      {"role": "user", "content": "Summarize the key risks in this deployment plan."}
    ],
    "reasoning_effort": "high",
    "stream": false
  }'

DeepSeek’s API documentation describes reasoning-effort settings including low, high, and max. Before a production evaluation, ask DeepSeek to confirm the model currently served by deepseek-v4-pro and the applicable billing rate. A successful API response confirms that a request was accepted; it does not, by itself, identify the served model. (DeepSeek API documentation; DeepSeek thinking-mode guide) (api-docs.deepseek.com)

DeepSeek API pricing for V4 Pro

DeepSeek’s API pricing page lists peak and off-peak prices per 1 million tokens for Pro-0813 and V4.1-Flash-1017. But the routing notice says requests to deepseek-v4-pro are billed at Flash rates during the stated interim period. The page’s Pro-0813 rates therefore should not be assumed to apply to that alias without confirmation of the current route and billing basis. (DeepSeek API pricing; DeepSeek routing notice) (deepseek.com)

Listed model and token category	Peak	Off-peak
Pro-0813: input, cache hit	$0.044	$0.022
Pro-0813: input, cache miss	$1.32	$0.66
Pro-0813: output	$3.96	$1.98
V4.1-Flash-1017: input, cache hit	$0.006	$0.003
V4.1-Flash-1017: input, cache miss	$0.30	$0.15
V4.1-Flash-1017: output	$1.20	$0.60

The pricing page says off-peak rates are half of peak rates. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. All other hours, including weekends and Chinese public holidays, are off-peak. DeepSeek notes that prices may change, so check its pricing page before budgeting or launch. (DeepSeek API pricing) (api-docs.deepseek.com)

Illustrative calculation from the listed rates: 1 million cache-miss input tokens and 100,000 output tokens would cost $1.32 + (0.1 × $3.96) = $1.716 at the listed Pro-0813 peak rates. At the listed V4.1-Flash-1017 peak rates, the same token volumes would cost $0.30 + (0.1 × $1.20) = $0.42. These are calculations from the rate table, not confirmation of what a deepseek-v4-pro request will serve or cost. Confirm the route and rate before using either as a production estimate. (deepseek.com)

A DeepInfra article lists rates for its own hosted offering and gives two different output-token prices in the same article. Those figures are not DeepSeek-direct pricing and do not establish how DeepSeek’s alias is routed. (DeepInfra’s V4 Pro overview)

Decision checklist

DeepSeek V4 Pro is a candidate to evaluate when your workload is text-based and could benefit from long context, reasoning controls, or tool use. Before committing:

1. Confirm the route and billing: ask DeepSeek whether deepseek-v4-pro currently serves Pro-0813 or V4.1-Flash, and which rate applies. The published routing notice says the Flash routing continues until V4.1-Pro launches; I found no official launch announcement in the sources checked.
2. Run a representative evaluation: measure quality, abstention, latency, tool success, and token use on your own tasks—and confirm which model served the evaluation.
3. Check modality needs: the API pricing page says Pro-0813 does not support vision; its V4.1-Flash-1017 listing does.
4. Budget by token category and schedule: distinguish cache-hit input, cache-miss input, and output; apply peak/off-peak prices only after confirming the effective route.
5. Recheck documentation at rollout: model listings, routing, and prices can change.

Bottom line: DeepSeek’s API documentation lists V4 Pro specifications and Pro-0813 rates. Its routing notice says requests to deepseek-v4-pro go to V4.1-Flash at Flash rates until V4.1-Pro launches. Because I found no official V4.1-Pro launch announcement in the sources checked, confirm the live route and billing with DeepSeek before treating a call as a V4 Pro evaluation or using Pro-0813 prices for a budget. (DeepSeek API pricing; DeepSeek routing notice) (deepseek.com)

Sources

Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient.

Models & Pricing | DeepSeek API Docs

Chat Completions API | DeepSeek API Docs

Your First API Call | DeepSeek API Docs

Thinking Mode | DeepSeek API Docs

DeepSeek-V4 Preview: Entering the Era of Affordable Million-Token Context

DeepSeek-V4-Pro GA Release | DeepSeek API Docs

DeepSeek V4 Pro: Model Overview, Features & Performance Guide