DeepSeek vs ChatGPT: the open-weight question, not just the cheaper one

DeepSeek and ChatGPT both answer questions in a chat window, and DeepSeek costs less. But the real difference is structural: DeepSeek publishes its model weights, so you can run it on infrastructure you control, while ChatGPT is only available as a hosted service from OpenAI. That changes who approves it, not just what it costs.

Most comparisons stop at price and benchmark scores. For anyone buying on behalf of a company, those are the two least durable facts in the whole discussion. Benchmarks turn over with every release, and list prices follow. What lasts is the deployment model, and on that point these two products are genuinely different animals.

Neither product appears in the Spotsaas catalogue, so there is nothing here we are selling you. Where a catalogue product is a useful reference point, it is linked.

These are two different buying decisions

Say “DeepSeek” and you could mean one of three things, and they carry different risk:

1. The consumer app and web chat, operated by DeepSeek itself. Free, quick to try, and the version almost everyone has actually used.
2. The DeepSeek API, also operated by DeepSeek, billed per token.
3. The open-weight models, published for download and run on your own servers or on a hosting provider you already have a contract with.

Say “ChatGPT” and you mean a hosted service from OpenAI, in a consumer tier, a business tier, or via the OpenAI API. There is no option three. If your security review requires that a model runs inside your own boundary, ChatGPT cannot satisfy it and DeepSeek can, and that single line decides more procurement conversations than any benchmark table.

	How you get access	Who operates it	What you are agreeing to
DeepSeek app or web chat	Sign up, free	DeepSeek	DeepSeek’s own published terms and privacy policy
DeepSeek API	API key, per-token billing	DeepSeek	DeepSeek’s own published terms
DeepSeek open weights	Download and self-host	You, or a host you choose	The model licence, plus your own hosting terms
ChatGPT consumer	Sign up, free or paid tier	OpenAI	OpenAI’s consumer terms
ChatGPT business or API	Contract or API key	OpenAI	OpenAI’s business terms and data processing agreement

DeepSeek vs ChatGPT cost: where the money actually shows up

On a per-token basis, DeepSeek’s API has been priced well below comparable frontier models, which is most of why it became a story at all. If your workload is high-volume and relatively simple, classification, extraction, summarising at scale, that gap is real money.

Three cost lines are easy to miss, though:

• Self-hosting is not free. Running an open-weight model well means GPU capacity, people who can keep it running, and an evaluation setup so you notice when quality drifts. For a small team, a hosted API is almost always cheaper once you count salaries.
• Seat pricing and token pricing are not comparable. ChatGPT’s paid consumer plans are per person per month. An API bill scales with usage. Comparing one figure against the other tells you nothing useful.
• Published prices move. Read xai.com, openai.com and deepseek.com before you build a model of your costs, because every number in every comparison article is out of date within a quarter.

For a sense of where per-seat pricing sits for assistants in this market, Perplexity AI starts at $20 and is rated 4.5 from 252 reviews on Spotsaas.

Data residency, stated plainly

DeepSeek is developed and operated by a company based in China. That is a fact about the vendor, not a verdict about the software, and it matters for exactly one reason: when you use the DeepSeek-operated app or API, your prompts are processed by a company subject to Chinese law, and the authoritative description of where that data is stored and how long it is kept is DeepSeek’s own published privacy policy. Read that document rather than anyone’s summary of it, including this one.

Three things a reviewer can establish from primary sources before approving anything:

• What the vendor’s own policy says about storage location, retention, and whether your inputs may be used for training. Both DeepSeek and OpenAI publish this. Both have separate, usually stricter, terms for paid and business use than for the free consumer product.
• Whether your sector or jurisdiction has issued guidance on the consumer app. Several regulated industries have, and your own counsel will know before you do.
• Whether the self-hosted route removes the concern. It largely does, and this is the pragmatic path most regulated teams take. Download the weights, run the model in a region you have already approved, and the residency question becomes a question about your own infrastructure, which you can answer.

The same discipline applies in the other direction. “Hosted in the United States” is also a jurisdiction, with its own legal reach, and a European buyer with GDPR obligations has to do the same reading on OpenAI’s terms that an American buyer does on DeepSeek’s. The honest framing is not safe versus unsafe. It is: know which laws apply, and write it down.

Where DeepSeek’s open weights change the calculus

Published weights buy you four things that no hosted service can offer: the model keeps working if the vendor changes direction, you can fine-tune it on your own material, you can pin a specific version so your outputs stay reproducible, and you can audit what the deployment actually does.

What you give up is everything built around a hosted product. Shared workspaces, administrative controls, retention settings, integrations, mobile apps, support when something breaks at 2am. ChatGPT’s business tiers exist precisely to supply that layer, and for most companies that layer is the product.

A reasonable split that a lot of teams land on: the hosted assistant for general staff use, an open-weight model behind your own API for the high-volume automated work where cost and control decide it.

DeepSeek vs ChatGPT for studying

For studying, the interesting part of DeepSeek is that its reasoning models show their working. Watching a model lay out its steps is genuinely good for learning, because you can see where the argument goes wrong rather than only that the answer was wrong. ChatGPT is stronger at the surrounding scaffolding: holding a long study plan, generating practice questions from material you paste in, and keeping explanations at a consistent level over weeks.

Whichever you use, check the reasoning against your source material. A fluent explanation that is subtly wrong is the hardest kind of error to catch, and both products produce them.

DeepSeek vs ChatGPT vs Gemini vs Claude

The four names that come up together differentiate less on answer quality than on shape, and shape is the thing you can still rely on in six months.

• DeepSeek is the open-weight option, so it is the only one you can run yourself.
• ChatGPT has the widest third-party ecosystem, so it is the most likely to already connect to something you use.
• Google Gemini is the one already inside Google Workspace, so if your documents and mail live there it reaches them with the least setup.
• Claude is the one teams most often pick for long documents and careful, cautious writing. Claude is in our catalogue, rated 4.44 from 9 reviews.

Is DeepSeek better than Claude? On cost and on control, for a team that can self-host, yes. On the hosted experience and on handling long, messy documents, most people who use both reach for Claude. They are answering different questions, which is why the comparison never settles.

Frequently asked questions

Where to go next

If you are evaluating assistants for a team, our chatbots and AI writing assistant categories cover the products built with administrative controls and business terms attached.

For the same comparison from a task-routing angle instead of a procurement one, read Grok vs ChatGPT.

Related reading

• ChatGPT Enterprise pricing: the contract-level numbers on the OpenAI side.
• Enterprise AI agents: platforms, pricing models and the security questions to ask.
• Claude Enterprise pricing: a third vendor’s terms for the same comparison.
• Claude vs ChatGPT: if the choice is narrowing to the two hosted options.