Deepseek Ai

• The company behind DeepSeek was founded in May 2023 by Liang Wenfeng, a Zhejiang University graduate who also co-founded High-Flyer, the quant hedge fund that owns DeepSeek, according to [TechTarge…
• The API is where DeepSeek's cost story is clearest.
• The data-handling question depends on the access path, and this is where most summaries go vague.
• Model choice follows the workload, not the brand name.
• The most cited difference is price-performance rather than feature parity. For instance, according to TechRadar, DeepSeek's R1 reasoning model briefly dethroned ChatGPT as the most downloaded free app on the U.S. iOS App Store in January 2025.

The company behind DeepSeek was founded in May 2023 by Liang Wenfeng, a Zhejiang University graduate who also co-founded High-Flyer, the quant hedge fund that owns DeepSeek, according to TechTarget. Its first model series began in November 2023, and DeepSeek-R1 followed on 20 January 2025, per Wikipedia). LinkedIn lists the company as privately held and headquartered in Hangzhou, with 51-200 employees. Funding and valuation have not been publicly disclosed, so treat any figure you see elsewhere as unverified.

Company details are as reported by the cited sources at the time of writing. According to TechRadar, DeepSeek's R1 claim sent Nvidia's stock tumbling 17% in a single session.

The API is where DeepSeek's cost story is clearest. TechTarget reports DeepSeek-V3.1 API pricing at $0.56 per million input tokens and $1.68 per million output tokens, against $1.25 input and $10 output for GPT-5. TechRadar reports that the V4 Flash API starts at $0.14 per million input tokens. LinkedIn notes a cut of more than 50% in API prices with V3.2-Exp. Chat access on the web and in apps is free to use, but the apps are proprietary, so the free tier is not a licence to embed the product in your own workflows.

Prices change; confirm current rates on DeepSeek's official pricing page before budgeting. According to TechRadar, DeepSeek had grown to roughly 97 million monthly active users by the time of its review.

The data-handling question depends on the access path, and this is where most summaries go vague. TechRadar identifies three main concerns:

• Data stored on Chinese servers
• Known security vulnerabilities
• Censorship on sensitive topics

Using the hosted chat or API sends prompts to infrastructure outside your control. Self-hosting the open weights moves inference inside your own environment, which removes the server-location concern but not the model's own behaviour or any vulnerabilities in your deployment. Custom-model traffic is also governed by the configured provider's policy, not DeepSeek's alone, so read each vendor's terms before routing customer or regulated data through it.

Verify against the current privacy policy for whichever access path you choose.

Model choice follows the workload, not the brand name. According to TechRadar, V4 Pro has 1.6 trillion total parameters with 49 billion active at inference, while V4 Flash has 284 billion total with 13 billion active, so Flash is the lighter, cheaper option. DeepSeek-R1 uses a Mixture-of-Experts architecture with 671 billion parameters and was positioned around reasoning, per RevEchat and Wikipedia. V3 remains the general-purpose predecessor. For API calls, deepseek-usa.ai, an independent guide not affiliated with DeepSeek, reports that the official documentation recommends the identifier 'deepseek-flash' for V4.1-Flash, so check the identifier against DeepSeek's own docs before deploying.

Parameter counts are as reported by the cited sources; confirm against DeepSeek's model documentation.

The most cited difference is price-performance rather than feature parity. DeepSeek-R1 briefly overtook ChatGPT as the most downloaded free app on the U.S. iOS App Store in January 2025, and by 27 January it had surpassed ChatGPT there, according to Wikipedia). TechRadar reports roughly 97 million monthly active users at the time of its review. Beyond downloads, the practical comparison comes down to where data lives, whether you need a closed-source vendor's ecosystem, and whether the open-weight licence matters to your team. Benchmarks change quickly, so judge each assistant on your own prompts rather than on headline rankings.

Other assistants are not covered by the sources here; add them only with a verified citation.

Yes, and the MIT License is the main reason engineering teams evaluate DeepSeek's open-weight models. According to TechRadar, DeepSeek's open-weight models are licensed under MIT, allowing teams to self-host and avoid per-token costs. According to Wikipedia, the R1, V3 and V4 models are all under the MIT License. However, the web and app products remain proprietary. Self-hosting therefore means running the open-weight models yourself, not the consumer chatbot apps.

The real cost shifts to hardware and operations. For example, DeepSeek-R1 has 671 billion parameters in a Mixture-of-Experts architecture, so it is not a laptop workload. Teams should plan for GPU capacity and an inference stack before committing. Specifically, confirm the hardware and software requirements for your chosen model in its official repository before deployment, since requirements differ across the R1, V3 and V4 families.

• Confirm the model family (R1, V3 or V4)
• Plan GPU capacity and an inference stack
• Check the official repository for requirements

The market reaction centred on the cost claim behind DeepSeek-R1. According to TechRadar, the DeepSeek-R1 release sent Nvidia's stock down 17% in a single session. According to TechTarget, the 27 January 2025 sell-off also hit Microsoft, Meta, Oracle and Broadcom.

According to RevEchat, DeepSeek-R1's developer cost was claimed at $6M, against an estimated $500M for OpenAI's comparable development.

The significance for buyers is the pressure DeepSeek-R1's pricing put on API rates across the market. TechTarget's comparison of DeepSeek-V3.1 API pricing against GPT-5 shows that gap directly: $0.56 versus $1.25 per million input tokens. That comparison does not settle the question of model quality.

Related guides

• Citation Submission Costs: What You Actually Pay Per Listing
• Local Citation Services: Build NAP Consistency That AI
• Citation Strategy for SaaS: Get Cited by AI Engines