DeepSeek profile

Developer access and other commercial offerings support the provider; a complete revenue breakdown is not verified here.

Annual sales are not publicly itemised for this business. Developer access and other commercial offerings support the provider; a complete revenue breakdown is not verified here.

Hosted developer access can generate usage income while open-weight models encourage adoption elsewhere. A model running on someone else’s servers does not automatically create hosted-chat revenue for DeepSeek.

Privately held; no public share price. DeepSeek develops AI models and hosted products. Open model weights allow other operators to run deployments, creating a distinction between the model developer and the company hosting a particular chat.

Developer access and deployment demand create commercial value around a widely distributed model ecosystem.

DeepSeek offers conversational and reasoning capabilities through hosted products and developer access. Downloadable model weights also allow separate deployments.

DeepSeek develops its model family; an independent deployment can be operated by someone other than DeepSeek.

The February 2026 policy describes prompts/uploads, account, technical and third-party data and use to train/improve models and technology. It lists rights, including an opt-out of training/technology-optimisation uses and chat-history controls. The effect of a particular request and legal retention still depend on the service and applicable law.

Hosted-service policies and model behaviour controls govern the official product; self-hosted deployments have their own operator and settings.

Distinguish the hosted chat from downloaded models when deciding where information will go.

15/25 points

DeepSeek’s published model availability should not be confused with the privacy properties of its hosted consumer service. A downloadable model and a conversation sent to an operated online service have different data boundaries.

10/25 points

The February 2026 policy describes prompts/uploads, account, technical and third-party data and use to train/improve models and technology. It lists rights, including an opt-out of training/technology-optimisation uses and chat-history controls. The effect of a particular request and legal retention still depend on the service and applicable law.

10/25 points

Model openness does not mean the hosted chat service has no data collection.

15/25 points

Distinguish the hosted chat from downloaded models when deciding where information will go.

Higher-scoring options for a comparable use:

• ChatGPT — 60/100 versus 50/100

  Comparable use case: general AI conversation

  General ai conversation: 60/100 compared with 50/100. A conversation is an unusually easy interface for turning one task into another.

  Sources (5)

  • ChatGPT — Trust evidencePublic primary documentation · accessed 18 October 2026
  • ChatGPT — Trust evidencePublic primary documentation · accessed 18 October 2026
  • ChatGPT — Trust evidencePublic primary documentation · accessed 18 October 2026
  • ChatGPT — official product pagePrimary product · accessed 18 October 2026
  • ChatGPT — Apple App StorePrimary store metadata · accessed 18 October 2026

• Claude — 60/100 versus 50/100

  Comparable use case: general AI conversation

  General ai conversation: 60/100 compared with 50/100. A polished draft can become an interactive tool without leaving the conversation.

  Sources (3)

  • Claude — Trust evidencePublic primary documentation · accessed 18 October 2026
  • Claude — official product pagePrimary product · accessed 18 October 2026
  • Claude by Anthropic — Apple App StorePrimary store metadata · accessed 18 October 2026

A useful answer or worked-through problem rewards another question.

An unfinished task and conversational follow-ups supply reasons to continue.

DeepThink and a simple chat interface make sophisticated model capabilities approachable.

Developer experiments and widely shared outputs spread interest in the models.

Saved conversations and an established workflow make changing assistants require context rebuilding.