Complete dining table, price history and the retired V3/R1 endpoints are on the brand new DeepSeek cost web page. 100 percent free tier, operates the modern V4 models, created by DeepSeek. I hold the model listing, API rates and you can log on procedures looked up against DeepSeek’s very own documents and you will print the new comment date for each webpage. To possess examples by using the Anthropic API style, delight make reference to Anthropic API.
- Sign in from the talk.deepseek.com — the official cam has a free of charge tier and operates the present day V4 patterns.
- Secrets, endpoints, design IDs, fast caching and password advice — in addition to just what changed if the dated speak and reasoner endpoints were retired.
- The newest mixed type in price includes per provider’s cached and uncached input rates at the cache hit rates; company as opposed to a published cached-type in speed try energized from the complete enter in speed.
- The sooner V3.step one and R1 patterns are still commonly used and you will unlock-weight, as well as the dated deepseek-chat and you can deepseek-reasoner API endpoints were resigned for the twenty four July 2026.
- Most recent DeepSeek habits and you can API costs for each and every million tokensModelBest forInput (cache miss)OutputGuideDeepSeek-V4.1-FlashFast relaxed model.
This is a different publication, maybe not the state provider. Before powering these examples, set the brand new DEEPSEEK_API_Trick environment varying on your terminal to your API trick. If you use products including Claude Password, GitHub Copilot, otherwise OpenCode, you should use DeepSeek while the backend model individually — zero password needed. The new DeepSeek API is supported by of a lot well-known AI broker and you will programming assistant products.
Dragon mega jackpot – See the DeepSeek Use Book to have info
The newest DeepSeek API uses a keen API format appropriate for OpenAI/Anthropic. The purchase price comparator in this article reveals just what same monthly work can cost you on each provider’s very own published cost, having a link to all rates credit. For each and every million tokens, V4-Thumb is actually $0.15 enter in (cache skip) and $0.6 productivity; V4-Professional are $0.66 input and you will $step 1.98 output. All of our demonstration isn’t the formal services and won’t work at V4. You can even is the new 100 percent free demonstration cam on this web site, and that works DeepSeek V3.step one for the people routes (shedding back to R1 whenever a course are hectic).
Live changelog away from model releases, rates changes and structure change. DeepSeek are a great Dragon mega jackpot Chinese AI lab whose high vocabulary patterns are strangely inexpensive to work at and you can open-pounds. That is a non-weight analogy, you can set the brand new stream factor to help you genuine to find stream effect.
Their latest fundamental design is actually DeepSeek-V4.1-Flash (API name deepseek-flash), fundamentally available since the 9 September 2026, with indigenous picture understanding manufactured in. DeepSeek AI is actually a Chinese AI look business one produces higher code designs. The state website is deepseek.com as well as the formal cam are speak.deepseek.com. The row links for the provider’s individual price webpage. Text message in addition to an experimental vision version.$0.15$0.6Read DeepSeek-V4-ProFlagship reasoning and coding model, to have more challenging multi-action work.$0.66$1.98Read DeepSeek’s very own speculative decryption bunch boats methods-alert schedulers and 2–3× token throughput on the H800s.
After you have received a keen API key, you can access the new DeepSeek design by using the after the analogy programs in the OpenAI API format.
Keys, endpoints, design IDs, prompt caching and you will code instances — as well as exactly what changed if dated chat and you can reasoner endpoints have been retired. Check in from the talk.deepseek.com — the official talk has a free tier and works the modern V4 designs. The brand new mixed input price brings together for each provider’s cached and you will uncached type in rate at the cache strike rates; business instead a circulated cached-type in price is billed at the complete type in speed. Latest DeepSeek designs and you may API prices for each and every million tokensModelBest forInput (cache miss)OutputGuideDeepSeek-V4.1-FlashFast casual model. The brand new organized services techniques prompts for the DeepSeek’s infrastructure in the China, numerous governments limitation they to the authoritative products, and you may notice-holding the fresh discover weights prevents sending analysis anyplace. The earlier V3.step 1 and you may R1 habits remain popular and open-pounds, as well as the old deepseek-chat and you will deepseek-reasoner API endpoints had been retired to your 24 July 2026.
Distilled versions match just one 24GB GPU; complete models you desire multi-GPU tools. To the published API costs DeepSeek is much lower per million tokens compared to GPT, Claude and Gemini flagships, and it is competitive to the programming and need criteria. The cost page have a complete table, cache-strike costs and you will speed history. Discover weights, resources you need, and you may step-by-step settings which have Ollama, vLLM and you may llama.cpp. See a good work, up coming see what the same volume will cost you on each provider’s published API rates.
