Skip to main content
Before integrating DeepSeek V4 through Rock API, verify the exact model route, current price, group, and time-band rule. The direct-provider schedule and a gateway’s configured billing expression are separate sources. Do not assume Rock API always charges a fixed percentage of the latest official rate. Public sources checked: November 2, 2026. Prices and configuration below are dated observations, not a future quote or an account-specific settlement.

​
Current official models and reference prices

DeepSeek’s official pricing page currently lists deepseek-flash served by DeepSeek-V4.1-Flash-1017 and deepseek-v4-pro served by DeepSeek-V4-Pro-0813. It says the legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp remain accepted by its own API, but their original models are retired and requests are served by the newer Flash version at its price. The following are direct DeepSeek reference rates, in USD per 1 million tokens. They are not Rock API effective prices. The official peak windows are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. Other hours, weekends, and those holidays are off-peak. Recheck the source before scheduling; this article does not establish when earlier prices became effective.

​
What Rock API’s public configuration shows

The unauthenticated public pricing endpoint lists deepseek-v4-pro and deepseek-v4-flash with dynamic tiered_expr billing and the openai endpoint type. It also exposes multiple group multipliers, including 0.7x: 0.7, 0.8x: 0.8, and 0.9x: 0.9. There are material differences from the current direct-provider page:
  • The public expression for deepseek-v4-flash contains input/output/cache-read coefficients 0.44 / 1.32 / 0.014 at peak and 0.22 / 0.66 / 0.007 off-peak, rather than the current official Flash rates above.
  • The public expressions for these two Rock API IDs test the UTC hour windows. They do not include weekday or Chinese-public-holiday conditions.
  • A gateway catalog entry does not establish that its legacy model name resolves to the same build as DeepSeek’s own API.
These observations describe published configuration, not an authenticated test of settlement or upstream routing. Confirm the route and billing details with the service before relying on official holiday scheduling or model-alias behavior. Do not mechanically multiply the official table by 0.7 and present the result as Rock API’s current quote.

​
Reconcile the selected group’s actual charge

Use the live catalog and pricing and billing to check the billing unit, currency, normalized input/cache/output categories, applicable group, and dynamic tier. A 0.7 group multiplier applies to the configured base cost if that group actually serves the request; it is not a universal discount against the current upstream list price. Avoid applying it twice to an effective rate. Top-up payment is separate from API consumption. For completed jobs, inspect the saved rule, matched tier, group, and settled usage. A current public configuration cannot reconstruct a historical bill, and a temporary reservation is not the final charge.

​
Integrate and schedule after validation

  1. Confirm the exact model ID, served version when disclosed, account/group access, and supported parameters.
  2. Configure credentials using authentication and send a small billable request using the quickstart.
  3. Test quality, tool arguments and structured output where supported, latency, errors, and usage records. Native-provider capabilities do not automatically carry through every gateway endpoint.
  4. For queued work, verify the route’s actual time rule and boundary handling; record UTC request times, retries, and matched tiers. Delay only jobs that tolerate waiting.
  5. Compare cost per accepted result before expanding production traffic.
Create a Rock API account when ready to verify account-specific conditions.