Important factors, endpoints, model IDs, punctual caching and password instances — and just what changed if the old speak and you can reasoner endpoints were resigned. Check in from the cam.deepseek.com — the official chat have a free of charge tier and you may works the present day V4 patterns. The brand new combined input rates includes per provider’s cached and you can uncached input speed at your cache strike price; company instead of a released cached-input price is energized during the full input rates. Latest DeepSeek designs and API costs per million tokensModelBest forInput (cache skip)OutputGuideDeepSeek-V4.1-FlashFast relaxed model. The new managed services process encourages to the DeepSeek’s structure within the China, numerous governing bodies restrict it to the official gadgets, and you may notice-holding the fresh open loads avoids sending investigation anyplace. The earlier V3.step 1 and you may R1 models continue to be popular and you can discover-weight, and also the old deepseek-chat and you may deepseek-reasoner API endpoints was resigned to your 24 July 2026.
DeepSeek Use is now in the developer preview to have broker utilize designers global. By the switching the brand new setting, you can use the newest OpenAI/Anthropic SDK or software programs suitable for the newest OpenAI/Anthropic API to view the newest DeepSeek API. Search / Get API is now offered, run on the within the-house search engine and help net look and you may bring.
Speak assistantRAG / document Q&ACoding agentBulk group History deepseek-cam and deepseek-reasoner aliases were retired; migrate so you can deepseek-v4-flash. V4-Flash and you may V4-Flash-Vision-Exp is retired; V4-Pro requests path to V4.1-Thumb from 14 September.
DeepSeek models and you may whatever they costs | this site
Once you have acquired a keen API key, you have this site access to the fresh DeepSeek model by using the after the example texts regarding the OpenAI API style. Distilled alternatives complement just one 24GB GPU; complete patterns you would like multiple-GPU methods. To your published API cost DeepSeek is actually much smaller for every million tokens compared to GPT, Claude and Gemini flagships, and is also aggressive for the programming and reason criteria. Our very own cost page provides an entire table, cache-hit costs and you may rates record. Open loads, methods you need, and you will action-by-step settings having Ollama, vLLM and you may llama.cpp. Come across a workload, then see what an identical frequency will set you back on each provider’s wrote API prices.
Heritage API aliases resigned — surcharge nonetheless not effective
Complete dining table, price background and also the resigned V3/R1 endpoints are on the new DeepSeek rates webpage. Free tier, works the present day V4 patterns, created by DeepSeek. I contain the model checklist, API costs and sign on tips looked facing DeepSeek’s own documents and you may printing the fresh comment day on each webpage. To possess instances with the Anthropic API format, excite make reference to Anthropic API.
Utilize the certified DeepSeek speak
- Discover a workload, then see what the same regularity can cost you for each provider’s authored API cost.
- Prior to powering this type of instances, place the brand new DEEPSEEK_API_Secret environment adjustable on the critical to the API secret.
- Open weights, methods you desire, and you can step-by-step options having Ollama, vLLM and you will llama.cpp.
- For the authored API costs DeepSeek try much lesser for each and every million tokens compared to GPT, Claude and you may Gemini flagships, and is also aggressive to your programming and reason standards.
- The prices web page have a complete desk, cache-struck cost and you can price record.
- This really is another book, maybe not the official service.
DeepSeek posts unlock weights to the GitHub and you may Hugging Face, and so they work at locally which have Ollama, vLLM, llama.cpp or LM Facility. This can be an independent book and you may people funding. Batch savings, long-context surcharges above a good provider’s endurance, 100 percent free sections and you will business sales are not provided.
Invoke The fresh Chat API
Their current main design try DeepSeek-V4.1-Flash (API label deepseek-flash), generally offered since the 9 September 2026, with indigenous image expertise produced in. DeepSeek AI are a Chinese AI research organization you to definitely generates highest vocabulary models. The official site is deepseek.com and the authoritative speak try speak.deepseek.com. The row links on the provider’s individual speed web page. Text as well as a fresh eyes variation.0.150.6Read DeepSeek-V4-ProFlagship reasoning and coding design, to have harder multiple-action work.0.661.98Read DeepSeek’s own speculative decoding stack ships tools-alert schedulers and you may 2–3× token throughput to your H800s.
Is actually the free demo chat
Real time changelog away from design releases, costs changes and you can structure transform. DeepSeek is actually an excellent Chinese AI laboratory whoever large code patterns is strangely cost effective to work with and you may discover-weight. That is a low-load analogy, you can place the newest weight factor to help you true to locate weight response.
DeepSeek publishes open loads on the GitHub and you can Hugging Face, and focus on locally with Ollama, vLLM, llama.cpp or LM Facility. This can be another book and you will community investment. Group deals, long-framework surcharges more than a great provider’s endurance, 100 percent free levels and you can company sales commonly provided.
DeepSeek Use is now inside designer examine for representative funnel designers international. By changing the brand new configuration, you need to use the fresh OpenAI/Anthropic SDK or software programs suitable for the brand new OpenAI/Anthropic API to get into the fresh DeepSeek API. Lookup / Get API is now available, powered by our very own within the-family google and you may help online search and you will get.