DeepSeek API pricing and developer access: token billing, granted balance, peak rates, and free Files API

DeepSeek’s developer platform is a metered API product even though consumer chat can be used without buying token credit. In September 2026, the API is built around V4.1 Flash and V4 Pro naming, a 1M-token context specification, account-level concurrency limits, and pricing that changes between peak and off-peak periods.

This page focuses only on developer economics: token charges, granted versus topped-up balance, model IDs, concurrency, and the cost effect of scheduling workloads outside peak hours.

··········

THE DEEPSEEK API IS METERED EVEN WHEN CONSUMER CHAT IS FREE

API usage is billed per million tokens. DeepSeek deducts charges from the developer account balance, using granted balance first when both granted and topped-up funds are available.

........

Model	Input cache hit	Input cache miss	Output
deepseek-flash off-peak	$0.003 / 1M	$0.15 / 1M	$0.60 / 1M
deepseek-flash peak	$0.006 / 1M	$0.30 / 1M	$1.20 / 1M
deepseek-v4-pro off-peak	$0.022 / 1M	$0.66 / 1M	$1.98 / 1M
deepseek-v4-pro peak	$0.044 / 1M	$1.32 / 1M	$3.96 / 1M

........

··········

PEAK AND OFF-PEAK PRICING CAN DOUBLE THE SAME WORKLOAD’S TOKEN BILL

DeepSeek’s current pricing page sets off-peak rates at half of peak rates. Peak windows are 01:00–04:00 UTC and 06:00–10:00 UTC, Monday through Friday; other periods are off-peak.

Data Studios calculation: a workload with 20 million uncached input tokens and 5 million output tokens costs about $6 on deepseek-flash off-peak and about $12 during peak hours. The same normalized workload is about $23.10 on the listed V4 Pro off-peak rates and $46.20 at peak.

........

Normalized workload	Off-peak	Peak	Peak premium
Flash: 20M input + 5M output	$6.00	$12.00	+$6.00
V4 Pro: 20M input + 5M output	$23.10	$46.20	+$23.10

........

··········

GRANTED BALANCE IS NOT THE SAME THING AS A UNIVERSAL FREE TRIAL

DeepSeek’s balance API distinguishes granted balance from topped-up balance. A developer account can therefore contain promotional or granted funds, but the existence of that balance type does not prove that every new account receives a fixed trial credit.

The practical rule is to check the account balance directly. If granted balance is zero, production API requests still require sufficient topped-up funds; an insufficient balance can return an HTTP 402 error.

··········

THE FILES API CAN BE FREE TO USE WHILE MODEL INFERENCE REMAINS BILLABLE

DeepSeek introduced a Files API for reusable image uploads and described the upload service itself as free to use. The model request that consumes the uploaded image is still tokenized and billed under the selected model’s inference pricing.

This is another reason to separate storage or upload mechanics from inference economics: a free file operation does not convert the model call into a free API request.

··········

CURRENT MODEL NAMES HAVE MOVED BEYOND DEEPSEEK-CHAT AND DEEPSEEK-REASONER

The legacy API names deepseek-chat and deepseek-reasoner were retired in July 2026 after a transition period where they mapped to V4 Flash modes. Current documentation uses deepseek-flash for V4.1 Flash and deepseek-v4-pro for the Pro line.

V4.1 Flash supports thinking and non-thinking modes under the current model architecture, so “chat model” and “reasoning model” are no longer best understood as two permanent product tiers.

··········

OFFICIAL PAGES CURRENTLY DISAGREE ON THE POST-SEPTEMBER V4 PRO ROUTING

DeepSeek’s September changelog says API service for V4 Pro continues after September 14, while the current pricing page also contains language indicating that requests to the V4 Pro model name may be routed to V4.1 Flash during the transition toward a future V4.1 Pro.

Because both statements are first-party and currently coexist, production users should verify the live pricing/model documentation at deployment time rather than hard-code an assumption about which backend is serving the V4 Pro identifier.

··········

CONCURRENCY LIMITS ARE A SEPARATE CAPACITY CONSTRAINT

DeepSeek currently lists account-level concurrency limits of 2,500 for deepseek-flash and 500 for deepseek-v4-pro. The limit is calculated across API keys belonging to the same account.

Requests above the concurrency ceiling can receive HTTP 429 responses. This is distinct from token billing: an account can have sufficient balance and still hit a concurrency limit.

··········

DATA STUDIOS COST MAP: SCHEDULING IS NOW A DIRECT API OPTIMIZATION

For batch jobs that do not need immediate execution, the peak/off-peak split creates a straightforward cost lever. Moving the same eligible Flash workload from peak to off-peak cuts the listed token price by 50% without changing the number of tokens processed.

For consumer access, free chat, files, and the difference between free use and API billing, see Data Studios’ DeepSeek free plans guide.

··········

FOLLOW US FOR MORE

··········

DATA STUDIOS

··········

datastudios.org