DeepSeek

DeepSeek offers a free web and mobile chat experience at chat.deepseek.com alongside a separate, usage-based API, and that combination is structurally different from how ChatGPT or Claude are priced. Those competitors gate their more capable models behind a flat monthly subscription (Plus, Pro), with the free tier limited to weaker or rate-capped access. DeepSeek instead makes its chat interface fully free with no Plus/Pro paywall and no usage cap on file uploads or conversation length, while charging separately, per-token, for anyone building on top of the API. For someone comparing "how much does this cost me," the practical answer depends entirely on which side of that split applies: a casual chat user pays nothing, while a developer integrating the API pays based on volume, similar to how OpenAI and Anthropic price their own APIs.

Model-wise, DeepSeek's flagship releases (the V4 family, alongside V3.2 and R1) are published as open-weight models on Hugging Face under the MIT license, one of the most permissive open-source licenses available. That means the weights can be downloaded, self-hosted, modified, and redistributed without licensing fees or usage restrictions from DeepSeek itself — a meaningfully different distribution model than closed competitors that only expose their models through a hosted API or app. Teams that want to avoid vendor lock-in, run inference on their own infrastructure, or audit a model's behavior directly have that option here in a way they don't with ChatGPT or Claude.

The genuinely relevant evaluation criterion that sets DeepSeek apart from most competitors on this list is jurisdiction. DeepSeek is a China-based company, and its official chat service processes and stores user data, including account details, prompt content, and uploaded files, on servers in China. Multiple governments (including Italy, Australia, Taiwan, and South Korea) have restricted or banned DeepSeek's hosted service on official devices in 2026, and a number of enterprises and public-sector organizations have added it to internal restricted-tools lists, citing concerns about legal data access by Chinese authorities. This is not a judgment on model quality — DeepSeek's benchmark performance is competitive with leading closed models — it's a data-governance and compliance question that matters most for regulated industries, government-adjacent work, or any team with contractual data-residency obligations. Anyone evaluating DeepSeek for anything beyond casual personal use should factor this in explicitly rather than treating it as identical to a US or EU-hosted alternative on that dimension.

For that reason, self-hosting the open-weight model on infrastructure the evaluator controls (rather than using the hosted chat.deepseek.com service or China-based API endpoint) is the option that sidesteps the jurisdiction question entirely, since data would never leave infrastructure the user already trusts. That path requires real infrastructure investment and technical setup, which is the same tradeoff any self-hosted open-weight model carries, DeepSeek included.

The practical decision splits into two separate questions: whether the free chat and low API pricing fit a given budget, and separately, whether the data-jurisdiction profile fits a given compliance or risk posture. Both are worth answering on their own terms rather than defaulting to either "it's free so it's fine" or "it's Chinese so it's disqualified" without checking the specifics against an actual use case.