DeepSeek vs ChatGPT (2026): DeepSeek Isn't the Cheap API Anymore, and the Data Question Still Decides

The short answer

As of October 23, 2026, DeepSeek is still the free chat app, but it is no longer the cheap API. OpenAI's GPT-6 Luna lists at $0.10 per million input tokens and $0.50 per million output; DeepSeek's deepseek-flash lists at $0.15 and $0.60 off-peak and $0.30 and $1.20 at peak. On a normal business workload, Luna is cheaper at both of DeepSeek's rates.

That flips the usual advice. A year ago the DeepSeek pitch was "nearly as good, a tenth of the price." Today the price edge is gone on the cheap tier, which leaves the question that always mattered more for a US business: DeepSeek's own privacy policy says it stores your data in the People's Republic of China.

The two side by side

Every API rate is per million tokens, from each vendor's own price page read today. The full DeepSeek rate card, including the peak-hour schedule, is on DeepSeek pricing, and every ChatGPT plan is on ChatGPT pricing.

The worked bill: 1,000 support conversations a month

Same workload I use on the DeepSeek pricing page: 1,000 customer conversations, about 3 million input tokens and 500,000 output tokens a month. List prices only, my arithmetic:

• GPT-6 Luna: 3 × $0.10 + 0.5 × $0.50 = $0.55 a month
• DeepSeek Flash, off-peak (the whole US workday): 3 × $0.15 + 0.5 × $0.60 = $0.75
• DeepSeek Flash, peak: 3 × $0.30 + 0.5 × $1.20 = $1.50
• DeepSeek V4 Pro, off-peak: 3 × $0.66 + 0.5 × $1.98 = $2.97
• GPT-6.1 Sol: 3 × $2.00 + 0.5 × $10.00 = $11.00

Now add caching, because a support bot sends the same system prompt and knowledge base every time. Say 80% of the input is a cache hit. DeepSeek Flash off-peak: 2.4M × $0.003 + 0.6M × $0.15 + output $0.30 = about $0.40. GPT-6 Luna: 2.4M × $0.01 + 0.6M × $0.10 + output $0.25 = about $0.33. DeepSeek's cache is cheaper per token, but Luna's lower output price still wins the month.

The honest reading of those numbers: at this volume, both are under a dollar. The model price stopped being the decision for small businesses a while ago. What costs real money is the build around it, the data review, and the hour someone spends cleaning up a bad answer. If you're choosing between the big US APIs instead, Claude API vs OpenAI API runs the same kind of math.

Where DeepSeek still wins

• The free chat app. For a person drafting, summarizing public material or practicing prompts, DeepSeek's app costs nothing and has no paid tier to upsell.
• The top-model price. V4 Pro at $0.66 / $1.98 off-peak is a fraction of GPT-6.1 Sol's $2 / $10. If a job needs a stronger model and the data is public, that gap is real.
• Open weights. DeepSeek's V4-generation models can be run on your own servers or a US cloud host, which takes DeepSeek's API, and its China storage, out of the picture. That's an engineering project, not a subscription.
• Search calls. $0.75 per 1,000 against OpenAI's $10 per 1,000, once DeepSeek starts billing on November 1.

The data question that settles it

DeepSeek's privacy policy (last updated February 10, 2026, read today) names a company with its registered address in China, says it stores personal data in the People's Republic of China, and says it uses data to train its models, with a right to opt out. For a US business that means: no customer records, no employee data, nothing under an NDA through DeepSeek's app or API. The full breakdown, including which governments restrict it, is on is DeepSeek safe.

ChatGPT has its own fine print. On Free, Go and Plus your chats can train models unless you turn it off. On Business, OpenAI says it doesn't train on business data by default. So the clean comparison for company work is DeepSeek's API against ChatGPT Business or OpenAI's API, and on that comparison the data terms point one way.

Which one I'd use

• Staff using AI at work: ChatGPT Business (or Claude Team). The seat price is the same $20 billed annually, and the data terms are the reason.
• A cheap API for a bot or an internal tool: GPT-6 Luna first. It's cheaper on list price, and your prompts go to a US company instead of one that stores data in China.
• Public-data jobs that need a strong model at a low price: DeepSeek V4 Pro is worth a test, off-peak, with nothing sensitive in the prompt.
• You want DeepSeek's models but not its servers: self-host the open weights, and budget for the engineering.

Today's micro-action: open your API bill and find your biggest model line. Multiply last month's input and output tokens by the rates in the table above. If the difference is under $20 a month, stop shopping models and spend that hour on the prompt. If you want the build, the data review and the team setup done in one pass, that's the AI implementation work I do.

Sources

DeepSeek, Models & Pricing (api-docs.deepseek.com/quick_start/pricing), including Search & Fetch pricing and the November 1, 2026 billing start, read October 23, 2026; DeepSeek Privacy Policy, "Last Update: Feb 10, 2026" (cdn.deepseek.com), read October 23, 2026; OpenAI, API pricing (developers.openai.com/api/docs/pricing), read October 23, 2026; OpenAI, ChatGPT plan pages (chatgpt.com/pricing, read October 19, 2026, and learn.chatgpt.com/docs/pricing, read October 23, 2026). Prices in USD. The monthly bills are my arithmetic from those list prices on a stated example workload; DeepSeek's chat app being free with no paid tier is as recorded on our DeepSeek pricing page.