# DeepSeek AI API in 2026: payment, cost and API key connection > Ways to top up the DeepSeek API, prices in USD, cache, checking your balance, and connecting a key to an agent; terms depend on the account and payment method. Source: https://ailoot.ai/blog/en/deepseek-api-payment/ Canonical: https://ailoot.ai/blog/en/deepseek-api-payment/ Language: en Publisher: Ailoot (https://ailoot.ai/) Translation: guide-135 Revision: sha256-8be4c71ecbb93aa8100a510269ac0cf465926587b19db0156bebafc03b57b87f There is no subscription; you pay for usage: 1 million input tokens for deepseek-flash off-peak cost \$0.15, and output tokens cost \$0.6. A typical month of agent use, with 10 million input tokens and 2 million output tokens, comes to \$2.7. If you have a payment method, the next steps come down to 2 things: top up the balance and add the key to the agent.
Key takeaway DeepSeek accepts payment through PayPal, bank cards, Alipay and WeChat Pay on the Top up page. The chat is free; you only need to pay for the API: a month of agent use on deepseek-flash off-peak costs about \$2.7. After topping up, add the key to an editor extension or to Claude Code.
## What you pay for in DeepSeek and what is free The chat on the DeepSeek website is free: the homepage explicitly says that chatting with the models is free. If you only need the chat, there is nothing to pay for. You need to pay for the API when the model runs inside your agent, bot or program. There is no API subscription. There is a balance: you top it up, and DeepSeek charges for each request using the formula “text volume × price”. Any granted balance you have is used first, followed by your top-up balance. Search results confuse these things: some describe a paid chat mode with automatic renewal, while others say there is no subscription at all. Use DeepSeek's own pages as your reference; the links are collected at the end. [Ailoot documentation: API, keys and connection setup](https://ailoot.ai/docs/). ## How much the DeepSeek API costs in 2026 Prices per 1 million tokens, checked on 8 October 2026 against DeepSeek's pricing page. Input is everything the agent sends to the model: instructions, files and conversation history. Output is what the model writes in response. Output is more expensive than input: 4 times as expensive for Flash and 3 times for Pro, so an agent that reads a lot and writes little costs less. | Model | Time | Input | Cached input | Model output | |-----------------|----------|--------|--------------|--------------| | deepseek-flash | off-peak | \$0.15 | \$0.003 | \$0.6 | | deepseek-flash | peak | \$0.3 | \$0.006 | \$1.2 | | deepseek-v4-pro | off-peak | \$0.66 | \$0.022 | \$1.98 | | deepseek-v4-pro | peak | \$1.32 | \$0.044 | \$3.96 | Peak hours in UTC+3 are from 04:00 to 07:00 and from 09:00 to 13:00, Monday through Friday. All other times, weekends and Chinese holidays are charged at the off-peak rate, which is half as much. If the agent runs in the evening in UTC+3 or on weekends, you pay less. Both models have a context window of 1 million tokens and an output limit of 384 thousand. If your agent settings still contain the old name deepseek-v4-flash, it is still accepted: requests go to Flash and are billed at the Flash rate. Now let's calculate a typical month of agent use: 10 million input tokens and 2 million output tokens. - deepseek-flash off-peak: 10 × \$0.15 + 2 × \$0.6 = \$2.7; at peak times, \$5.4. - deepseek-v4-pro off-peak: \$6.6 + \$3.96 = \$10.56; at peak times, \$21.12. I showed how to estimate your own 10 million input tokens [in a separate article](https://ailoot.ai/blog/en/api-token-usage-calculation/). Caching reduces the bill the most. Each time, the agent sends the model the same long instructions and project description. Repeated content the model has already seen is served from the cache, and on Flash it costs 50 times less: \$0.003 rather than \$0.15 per 1 million tokens. The longer the agent's repeated context, the more noticeable the difference. If you see Flash prices of \$0.14 for input and \$0.28 for output anywhere, those are old rates and no longer apply. Check the pricing page before topping up; DeepSeek itself recommends doing so. If you build your own projects with AI agents, check the connection settings against the documentation. ## If your bank or payment gateway declined the card The reason a particular card is declined is determined by your bank or payment gateway. DeepSeek does not publish the reason for a refusal. Trying the same card over and over is therefore pointless; it is better to move straight to another method from the list. A VPN has no effect on payment either. It changes the connection address, but the bank that issued the card remains the same. A card that was rejected without a VPN will be rejected with one too. ## Which of your own payment methods to use Any method on DeepSeek's list will do, provided it is your own. 1. **Your own PayPal.** If you have a working PayPal account, use it to top up. 2. **Your own Alipay or WeChat Pay wallet.** If you already work with China and have a wallet, pay with it. Whichever method you use, check 3 things before paying: - **Whose account and key you are using.** Only your own. Anyone who knows your key can spend your balance. - **Whose name the card or wallet is in.** Only yours. - **How much you will pay in total.** Your bank or payment system may charge its own fee on top of the top-up amount; check it before paying. I have [a separate article](https://ailoot.ai/blog/en/ai-service-card-declined/) on paying for overseas AI services when your card is rejected. ## How to top up your DeepSeek balance step by step 1. Sign in to platform.deepseek.com with your own account. 2. Open the Top up page: . 3. Choose a payment method: PayPal, bank card, Alipay or WeChat Pay. 4. Use the amount from the monthly calculation for your model. 5. After paying, check the payment’s transaction status and the top-up balance in your account dashboard, separately from any granted balance. 6. Make the first request from the agent and check the response against the table below. A separate point about the amount: there is no reason to add extra balance “just in case”. DeepSeek itself recommends topping up according to actual usage. If you later want to get the remaining balance back, only the entire unused balance will be refunded in one payment, minus fees. For an agent using deepseek-flash off-peak, a month based on the calculation above costs \$2.7; for deepseek-v4-pro, it costs \$10.56. Start with that, then check the dashboard after a month to see how much was actually used. To confirm the payment, check the transaction status and the change in your top-up balance in the account dashboard. A successful first request separately checks the key and API access; it may be paid from an existing granted balance, so it does not prove that this top-up arrived. If the request fails, use the response to narrow the issue: | What you see | What it means | What to do | |----|----|----| | Card declined | The bank or gateway did not authorize the payment | Choose another method from the list; do not retry the same card | | Error 402 Insufficient Balance | The balance has not been topped up or has run out | Check the balance and top it up on the Top up page | | Error 401 | Incorrect key | Copy the key again and paste it without spaces | | Error 429 | Too many requests in a short time | Pause and send requests less frequently | | Money was debited, but no balance appeared | The payment has not yet reached the account dashboard | Do not pay a second time; check the dashboard and your bank statement | ## How to connect a DeepSeek AI API key to an agent There are 2 routes. The first is an editor extension that already lists DeepSeek as a provider. Add your DeepSeek API key, and that is it. You then work with the model directly in the editor. The second is Claude Code and other agents. DeepSeek can respond in the same format as Anthropic, so in the agent settings you change the address to and specify your own key. In Claude Code, these are 2 environment settings, shown exactly this way on DeepSeek's page: ```bash export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic export ANTHROPIC_API_KEY=your_key_from_platform.deepseek.com ``` For programs that use the OpenAI format, the address is . DeepSeek itself says that Claude Code, GitHub Copilot and OpenCode work with it without code. The model name is deepseek-flash or deepseek-v4-pro, based on the calculation in the pricing table. Store the key in environment settings (environment variables) and do not put it in the project text; otherwise, it will go with the files wherever you publish them. Then send the agent a short request. If it responds, the key works. If you get an error, find it in the table in the previous section. ## Incorrect claims about DeepSeek **Claim: your top-up balance expires.** DeepSeek's FAQ explicitly says: Top-up balances do not expire. Your top-up balance does not expire. **Claim: you cannot get your money back.** You can. The FAQ has a Refunds button, and under the platform's terms, the entire unused balance is refunded in one payment, minus fees. There are no partial refunds, and spent funds are not refunded. ## Is DeepSeek the one worth paying for? I consider it the best of the Chinese models right now, the leader, especially in terms of quality. It has a context window of 1 million. It is good enough, quite good. The very first simple step to save money is to switch the agent's model from Sonnet to DeepSeek version 4. That alone will cut the price by about 5 times. I do not remember exactly, but it is roughly that. Unfortunately, DeepSeek cannot complete a long task with many steps on its own. I tried it, and it breaks down along the way. This is my subjective assessment; you can only find out through testing. DeepSeek is good for high-volume routine agent work where price is what matters. Give a long task with many steps to a stronger model or split it into short pieces. I compared how DeepSeek's bill stacks up against Claude Code in a separate article. Compare the model your agent uses and its actual monthly spend.
### Frequently asked questions Does the token price differ depending on the payment method? DeepSeek charges the same price for all methods. Your bank's or payment system's fee is added on top, and fees are also deducted when the remaining balance is refunded. How do I top up my DeepSeek API balance by exactly the amount I need? DeepSeek recommends topping up according to actual usage. Calculate a month using the pricing table for your model, top up by that amount, then check the charges in your dashboard after a month and adjust the next top-up.
### Sources - [DeepSeek: Models & Pricing](https://api-docs.deepseek.com/quick_start/pricing) — deepseek-flash and deepseek-v4-pro prices, peak hours, old model names, context of 1 million tokens and the order in which balances are charged - [DeepSeek: FAQ](https://static.deepseek.com/faq/index.html?lang=en) — payment methods on the Top up page, the balance does not expire, refunds through Refunds - [DeepSeek: Open Platform Terms of Service](https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html) — refund of the unused balance in one payment, minus fees - [DeepSeek: Error Codes](https://api-docs.deepseek.com/quick_start/error_codes) — errors 401, 402 and 429 - [DeepSeek: Anthropic API](https://api-docs.deepseek.com/guides/anthropic_api) — the address for Claude Code and agents - [DeepSeek: API FAQ](https://api-docs.deepseek.com/quick_start/faq) — Claude Code, GitHub Copilot and OpenCode work with DeepSeek without code - [DeepSeek: homepage](https://www.deepseek.com/en/) — chatting with the models is free
If you need access to AI models through Ailoot, explore [Ailoot](https://ailoot.ai/). ## DeepSeek AI API key questions ### Does a successful API response prove that my top-up arrived? No. An existing granted balance may fund the request first. Confirm the payment through the transaction status and the top-up balance in your dashboard; use a small request separately to check the key and API connection. ### What should I do if the bank shows a debit but the dashboard does not show the top-up? Do not pay again immediately. Compare the transaction status, top-up balance, and bank statement, and resolve that payment’s status before attempting another top-up. ## Ailoot · The company and its project portfolio Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control. The company and its project portfolio: millions of customers; monthly sales in the billions. Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.