DeepSeekCompanyDeepSeekAI company behind the DeepSeek family of reasoning models.DeepSeek APIOpen companyAI Model API
DeepSeek API
DeepSeek's first-party API serves deepseek-flash (DeepSeek-V4.1-Flash-1017) and deepseek-v4-pro with a 1M-token context, in OpenAI- and Anthropic-compatible formats. It bills per million tokens from a prepaid balance, with cheap cache hits and half-price off-peak hours, and stores data in the People's Republic of China.
The 30-Second Take
Pick the DeepSeek API if you choose models on price per task and your data may be stored in China.
Practitioners praise it above all for cheap cache hits and half-price off-peak hours. Flash costs US$0.30 per million cache-miss input tokens at peak, half that off-peak.
- Best fit
- Developers and product teams whose data may leave their jurisdiction and who choose models on price per task.
- Main constraint
- Data stored in the People's Republic of China under PRC law, with no SSO or audit log documented.
- First move
- Clear the privacy policy and terms with legal, then run a week of real traffic off-peak and measure cache hits.
Why
Watch the data terms: DeepSeek stores personal data in the People's Republic of China and trains its models on it. US federal civilian agencies may not use it, and practitioners report empty answers and leaked tool-call markup.
37 practitioners, 48 sources reviewed, Updated Oct 31, 2026How this is measured
- Operating stack
- AI Application Platform
Models
Which models does DeepSeek API serve, and where is each strong?
Each model's page charts its public benchmark scores beside any other model you layer on it.
- DeepSeek V4.1 FlashDeepSeek's smallest model in its V4.1 architecture family, with native visual understanding, served on the DeepSeek API as deepseek-flash. Since October 17, 2026 that name serves the DeepSeek-V4.1-Flash-1017 build. Weights under the MIT license.
- DeepSeek V4 ProDeepSeek's largest V4 model, a mixture-of-experts model with 1.6T parameters and 49B active and a 1M-token context, served on the DeepSeek API as deepseek-v4-pro with text input only. The 0813 build is its generally available release, superseding the April preview. Weights under the MIT license.
Comparisons
How does DeepSeek API compare with the alternatives?
DeepSeek API vs OpenRouter
- AI Model APIs
Buy from DeepSeek for its cache discounts and first-party quality when data may be stored in China. Route the open weights through OpenRouter when you need Zero Data Retention routing or hosts outside China, and test each host.
DeepSeek API vs Claude API
- AI Model APIs
Pick DeepSeek for price per task on long agent sessions when its data terms are acceptable. Pick the Claude API for zero data retention by request, US-only inference, and cloud routes.
Operator Insights
What are teams learning from building on DeepSeek API and running it in production?
Zaeema NashathSenior Software EngineerLanka Data Foundation
Lanka Data Foundation, which publishes Sri Lankan public data at opendata.lk, archives government gazettes with its open-source gztarchiver.
GitHub (LDFLK/gztarchiver), published Read the issue (opens GitHub (LDFLK/gztarchiver) in a new tab)
Observation
Lanka Data Foundation uses the DeepSeek API to classify Sri Lankan government gazettes, and an HTTP 429 burst can leave a large part of a batch uncategorised until the next run.
- Takeaway
- Throttle batch jobs on the DeepSeek API and retry 429s with backoff, because an unthrottled burst can fail a large part of a batch.
- What happened
- gztarchiver calls DeepSeek once per gazette, back to back and with no throttling, so a full year can mean hundreds of requests in quick succession to the DeepSeek endpoint. When the API returns HTTP 429, the document is marked NOT CATEGORISED and logged as a failed classification, and it is retried only on the next run. The team proposes a requests-per-minute limit and exponential backoff that honours Retry-After before a document is marked failed.
- Public engineering write-up
- Failure mode described
Deployment details
Lanka Data Foundation, which publishes Sri Lankan public data at opendata.lk, archives government gazettes with its open-source gztarchiver.
- Public engineering write-up
- Failure mode described
- Team
- Lanka Data Foundation
- Workload
- Classifying Sri Lankan government gazettes, hundreds of requests for a full year
- Product version
- DeepSeek's API, called through a WSO2 Bijira gateway
Dan EwartVice President of Information Technology and Chief Information OfficerUniversity of Idaho
The University of Idaho's networks and university-managed computers, under a memo to all employees from the CIO and Chris Nomura, Vice President for Research and Economic Development.
University of Idaho (support.uidaho.edu), published Read the memo (opens University of Idaho (support.uidaho.edu) in a new tab)
Observation
The University of Idaho dropped DeepSeek: on Aug 7, 2025 it blocked DeepSeek's web interface and API on university networks and told employees to stop using it, for Department of Energy research compliance.
- Takeaway
- If your team does federally funded research, clear the DeepSeek API with research compliance before building on it, as the University of Idaho blocked it on its networks.
- What happened
- The memo, dated Aug 7, 2025, says access to DeepSeek AI services was restricted across university networks to comply with Department of Energy and other expected research and security requirements, citing federal DOE guidance that calls for heightened scrutiny of AI platforms that may pose data risks to federally funded research. DeepSeek web interface and API access is blocked on university networks, and DeepSeek is blocked on university-managed computers. Existing use must move to alternative AI applications at once, with limited exceptions where explicitly authorized.
- Public organization write-up
- Signed university memo
- Scope and reason stated
Deployment details
The University of Idaho's networks and university-managed computers, under a memo to all employees from the CIO and Chris Nomura, Vice President for Research and Economic Development.
- Public organization write-up
- Signed university memo
- Scope and reason stated
- Team
- University of Idaho employees
- Workload
- Any use of DeepSeek's web interface and API on university networks and university-managed computers
- Product version
- DeepSeek web interface and API
Buying and deployment
What a team building on it should verify before production traffic grows.
Sourced from DeepSeek documentation
Signals
What is the current evidence saying about DeepSeek API?
A current read of what developers and teams report publicly about building on DeepSeek's own API, with each finding linked to its sources.
Practitioner sentiment, all practitioners
+57 Strong positive net
of 39 sources that take a side, each author counted once.
How this figure is madeTeams and leaders have 0 accounts; GILD prints their sentiment from 20.
Current sentiment
Positive, mixed and negative sources
The score counts them as 27 positive, 4 mixed and 6 negative votes, one per author.
Across 43 sources the read is positive. Of the 39 that state a view, 29 are positive, 4 mixed, and 6 negative. Price per task and cache hits draw praise in 25 sources, and output bugs, latency, and errors are raised in 5.
Run a pilot on your own prompts to measure output quality and cost per task.
Evidence field. All 48 sources, one mark each.
Arrange by
Each mark is one source: positive, mixed, negative, no opinion stated, or documentation (not counted).
| Group | Sources | Positive | Mixed | Negative | No opinion stated and not counted in the score | Documentation, checked and not counted |
|---|---|---|---|---|---|---|
| Positive | 29 | 29 | 0 | 0 | 0 | 0 |
| Mixed | 4 | 0 | 4 | 0 | 0 | 0 |
| Negative | 6 | 0 | 0 | 6 | 0 | 0 |
| Not counted | 4 | 0 | 0 | 0 | 4 | 0 |
| Documentation | 5 | 0 | 0 | 0 | 0 | 5 |
| Quarter | Sources |
|---|---|
| Q2 2026 | 12 |
| Q3 2026 | 16 |
| Q4 2026 | 15 |
| Documentation, checked | 5 |
By sentiment
5 documentation pages check facts and are not counted.
Hover a mark for a quick look, click it or press Enter to open its source.Tap a mark to open its source, then step to its neighbors with Previous and Next.
What practitioners praise
Each mark is a source on it; filled marks praise it.
Most common concerns
Each mark is a source on it; filled marks raise it.
All 48 sources behind these findings
- Models and pricingDeepSeek (Official), Product documentation, checked Oct 28, 2026
- Rate limitDeepSeek (Official), Product documentation, checked Oct 28, 2026
- Your first API callDeepSeek (Official), Product documentation, checked Oct 28, 2026
- Privacy policyDeepSeek (Official), Product documentation, checked Oct 28, 2026
- Open Platform terms of serviceDeepSeek (Official), Product documentation, checked Oct 28, 2026
- Intermittent EPROTO: The TLS handshake failed. Does the server speak TLS on this port?Joseph Cake, AI builder using Kilo Code (self-described: Kilo Code agent sessions against api.deepseek.com), GitHub issue, published Oct 24, 2026
- Comment on: Kimi K3.1 Pro and K3.1 Pro HighspeedSwellJoe, Hacker News commenter (self-described: Personal API use), Hacker News comment, published Oct 24, 2026
- A vibe-coded MITM test proxy built for about $0.30 of DeepSeek API spendaftbit, software developer in Ohio, at Atomic Canyon in 2025 (self-described: Building a test tool), Hacker News comment, published Oct 24, 2026
- Comment on: MiniMax M3.1 FlashTheArcane, Hacker News commenter (self-described: Pays for both the DeepSeek platform and a US$100 Claude plan), Hacker News comment, published Oct 20, 2026
- Comment on: DeepInfer: High-performance inference framework for DeepSeek modelsVonLuderitz, Hacker News commenter (self-described: Personal use after leaving a Claude Max plan), Hacker News comment, published Oct 20, 2026
- A platform.deepseek.com account beside a $100 a month Claude subscription, and DS Flash used mostTheArcane, autonomous driving perception research engineer (self-described: Holds both a DeepSeek platform account and a Claude subscription), Hacker News comment, published Oct 18, 2026
- Comment on: DeepSeek-v4.1-Flash-1017computerex, Hacker News commenter (self-described: Comparing a US$20 OpenAI plan with API spend), Hacker News comment, published Oct 17, 2026
- [Experimental Feature Feedback] Routing DeepSeek V4.1 Flash Requests to DeepSeek's Own APIDuskriver, maintainer of the dsh-opencode-go plugin (self-described: OpenCode Go requests for DeepSeek V4.1 Flash), GitHub issue, published Oct 16, 2026
- Comment on: The cost of getting a job today63stack, Hacker News commenter (self-described: Personal API use), Hacker News comment, published Oct 15, 2026
- Comment on: Why isn't the industry freaking out about DeepSeek 4.1 Flash?WiSaGaN, Hacker News commenter (self-described: Everyday personal tasks from the pi harness), Hacker News comment, published Oct 9, 2026
- A simple loop calls the DeepSeek API for JSON that invokes native Nokia functionsanupray, Software Engineer (self-described: A small loop that invokes native Nokia functions), Hacker News comment, published Oct 9, 2026
- Comment on: Why isn't the industry freaking out about DeepSeek 4.1 Flash?try-working, Hacker News commenter (self-described: Heavy agentic software work on V4.1 Flash), Hacker News comment, published Oct 8, 2026
- Add first-class DeepSeek provider (and track other major model providers)LaneBucher, GitHub user (self-described: Lumen's model adapters), GitHub issue, published Oct 3, 2026
- Comment on: One month coding with GLM 5.3 Flashbenjiro29, Hacker News commenter (self-described: Long coding tasks across hosts), Hacker News comment, published Oct 2, 2026
- Comment on: One month coding with GLM 5.3 Flasheli, Hacker News commenter (self-described: Choosing a host for DeepSeek models), Hacker News comment, published Oct 2, 2026
- [BUG] deepseek-flash emits internal DSML tool-call markup in content when tools array is omitted but history carries tool_callsDoll-sensei, operator of a QQ bot on the DeepSeek API (self-described: A production QQ bot on the API), GitHub issue, published Sep 30, 2026
- 1.8bn tokens in a week bought from DeepSeek, 99.2% of them cache hits, for US$18.13taylorfinley, serial builder of things (self-described: A week of tokens bought from DeepSeek itself), Hacker News comment, published Sep 29, 2026
- Weeks without topping up a DeepSeek balance, against weekly OpenAI top-upsdools, runs BenkoPhone (self-described: Planning and implementation on prepaid balances), Hacker News comment, published Sep 29, 2026
- [BUG] deepseek-v4-pro and v4.1-flash: finish_reason "stop" with the answer in reasoning_content and content emptyAmiyanshu-web, developer replaying production conversations (self-described: Streaming with thinking enabled on api.deepseek.com), GitHub issue, published Sep 24, 2026
- I built Codex Switcher - Open source Mac menu bar app to use Deepseek with CodexSystem_Independent (maskaravivek on GitHub), builder of Codex Switcher (self-described: Codex pointed at DeepSeek after a subscription ran out), Reddit post, published Sep 19, 2026
- Comment on: Ask HN: What default model do you use and why?tiluha, Hacker News commenter (self-described: V4.1 Flash agents with a frontier advisor model), Hacker News comment, published Sep 12, 2026
- DeepSeek as the default model on ChatGPT CodexEkky Armandi, full-stack Python developer (DEV profile) (self-described: Codex configured to use DeepSeek Flash through the official API), Developer blog post, DEV Community, published Sep 12, 2026
- Comment on: DeepSeek launching v4.1 flash cheaper and more capable than v4 promattmcal, Hacker News commenter (self-described: OpenCode on the DeepSeek API, OpenRouter, and Vercel AI Gateway), Hacker News comment, published Sep 9, 2026
- Comment on: DeepSeek V4 Pro 0813irthomasthomas, Hacker News commenter (self-described: Choosing between the DeepSeek API and OpenRouter), Hacker News comment, published Aug 13, 2026
- Comment on: DeepSeek V4 Pro 0813nchmy, Hacker News commenter (self-described: Long chats on the official API), Hacker News comment, published Aug 13, 2026
- deepseek-v4-flash terminates prematurely in agent tool-call loops via OpenAI-compatible APILady-Lin, developer running a coding agent on the API (self-described: A coding agent's execution model on api.deepseek.com), GitHub issue, published Aug 9, 2026
- Uses the DeepSeek API, pays peanuts, and is very satisfiedstavros, founder of Stochastic Technologies (self-described: Personal API use), Hacker News comment, published Aug 7, 2026
- Comment on: DeepSeek planning to significantly raise pricesshortformblog, Hacker News commenter (self-described: Personal use of the API), Hacker News comment, published Aug 6, 2026
- The email DeepSeek sent its API users about a planned, significant price risemittermayr, self-employed, formerly a Microsoft Product Planner (self-described: An email to DeepSeek API users), Hacker News comment, published Aug 6, 2026
- $5 a month going directly through DeepSeek's API all dayBayart, SRE (self-described: Light and medium tasks all day long), Hacker News comment, published Aug 3, 2026
- The DeepSeek API for all coding at home, through OpenCode or Pi, for $5 over six monthseulers_secret, works on embedded Linux for avionics, formerly at Intel for more than 10 years (self-described: All coding at home), Hacker News comment, published Jul 25, 2026
- 287 million tokens in a couple of days on DeepSeek's API, for US$5.77lionkor, software engineer in the electrical power industry (self-described: Calling DeepSeek's API directly), Hacker News comment, published Jun 25, 2026
- Happy buying tokens directly from DeepSeekmark_l_watson, author of 20 books and a practitioner specializing in artificial intelligence (self-described: Tokens bought directly from DeepSeek), Hacker News comment, published Jun 25, 2026
- Anyone else find managing separate API keys/billing across LLM providers annoying?NoMaterialMan, founder building a product on the Claude, GPT and DeepSeek APIs (self-described: A product calling Claude, GPT and DeepSeek through each provider's own API), Reddit post, published Jun 20, 2026
- DeepSeek's platform API with V4 Flash in OpenCode, more than acceptable for maybe $10 over a couple of weeksatreids, Software Engineer in Scotland (self-described: DeepSeek's platform API called directly from a harness), Hacker News comment, published Jun 13, 2026
- 15 cents this month on the DeepSeek API with a home-built coding agent, and three apps delivered to clientsKuyawa, programmer (self-described: Client app work with a home-built coding agent), Hacker News comment, published Jun 13, 2026
- DeepSeek's Anthropic endpoint for bigger reading jobs near the limitCJefferson (Christopher Jefferson), Professor in the School of Computer Science at the University of St Andrews (ANTHROPIC_BASE_URL pointed at DeepSeek near the limit), Hacker News comment, published Jun 8, 2026
- The DeepSeek API caches raw API calls well, so it is hard to spend a lot fastSwellJoe, works on Virtualmin (self-described: Raw API calls from a simple agent harness), Hacker News comment, published Jun 8, 2026
- The DeepSeek dashboard shows around 100 million tokens a day, almost all of it cacheKronisLV, software engineer from Latvia (self-described: DeepSeek token spend set against an Anthropic Max subscription), Hacker News comment, published Jun 8, 2026
- A custom terminal coding agent with file tools only, connected to the DeepSeek APIKuyawa, programmer (self-described: Building apps from a prompt with a home-built terminal agent), Hacker News comment, published May 30, 2026
- DeepSeek API credits never expire, and the prices are ridiculously low for the qualitysourcecodeplz, full-stack engineer (self-described: Prepaid DeepSeek API credits), Hacker News comment, published May 27, 2026
- US$1.15 used after two full weekends on a 2 USD DeepSeek API top-upvinhnx, iOS & Applied AI Engineer (self-described: A DeepSeek API key with a 2 USD top-up), Hacker News comment, published May 25, 2026
- 3.4 Billion Tokens for $27: A Real DeepSeek Cost Analysisxulingfeng, QA engineer of 15 years building an AI test framework (self-described: A month of DeepSeek bills), Developer blog post, DEV Community, published May 25, 2026
May 2026 to Oct 2026.
43 sources from practitioners, by month and sentiment.
Where It Fits
Where does DeepSeek API belong in the stack?
DeepSeek API is Core across Model, supports Build and Data, and touches Evaluate and Operate.
- Business area
- Product & Engineering
- Function
- Software Development
- Operating stack
- AI Application Platform
- Primary role
- Low-Cost Model API
Model
Model APIs: Core
Model APIs, core. This is the API's primary role: deepseek-flash (DeepSeek-V4.1-Flash-1017) and deepseek-v4-pro (DeepSeek-V4-Pro-0813), each with a 1M-token context and up to 384K output, billed per 1M tokens with half-price off-peak hours.
Model gateway: Touches
Model gateway, touches. The API serves only DeepSeek's models from DeepSeek's own servers. Gateways such as OpenRouter and LiteLLM call it, or other hosts of the open weights, and handle routing and failover.
Build
Orchestration: Touches
Orchestration, touches. Tool calls, JSON output, and the Responses and Anthropic Messages formats let agent frameworks and coding agents such as Claude Code, Codex, and OpenCode run on it. The framework controls the agent's steps.
Tool use: Supports
Tool use, supports. Tool calls are supported on both models, and the Search and Fetch APIs give agents web search and page fetches at US$0.75 per 1,000 successful requests each. The tools an agent calls otherwise run elsewhere.
Data
Retrieval: Supports
Retrieval, supports. Search and Fetch ground answers in the web, and the 1M-token context holds long documents. Document stores and search over company data stay with a retrieval product.
Evaluate
Evaluation: Touches
Evaluation, touches. The API offers no evaluation tooling, so a team scores outputs in an evaluation product.
Operate
Observability: Touches
Observability, touches. The API returns token usage per call, while tracing and cost dashboards stay with an observability tool or gateway.
Cloud deployment: Touches
Cloud deployment, touches. Capacity is a per-account concurrency limit of 2,500 requests for deepseek-flash and 500 for deepseek-v4-pro, raised on request at no cost. DeepSeek's privacy policy places data storage in the People's Republic of China, and the open weights can be hosted elsewhere.
Guardrails: Touches
Guardrails, touches. Screening prompts and outputs for injection, jailbreaks and sensitive data runs in a guardrail product or gateway placed in front of the API.
Market mapThe layers DeepSeek API covers, and who else owns each one.
Model APIs
Tool use & integrations
Retrieval & grounding
Model gateways & routing
- LiteLLM
- OpenRouter
- Portkey
- Vercel AI Gateway
Orchestration & agent frameworks
4 more layers on the full map. Core, Supports and Touches describe coverage. They say nothing about quality.