Tips, endpoints, design IDs, quick caching and you can password advice — as well as what changed when the dated cam and you can reasoner endpoints had been resigned. Register in the cam.deepseek.com — the official talk features a free of charge tier and you can operates the present day V4 patterns. The brand new mixed type in rate combines for each provider’s cached and you may uncached type in rates at your cache struck rates; team instead of a circulated cached-enter in speed are billed at the complete enter in price. Newest DeepSeek patterns and API rates for each million tokensModelBest forInput (cache miss)OutputGuideDeepSeek-V4.1-FlashFast casual model. The fresh organized solution processes encourages on the DeepSeek’s structure in the Asia, several governing bodies restrict they on the authoritative devices, and you can self-holding the newest unlock loads stops sending investigation anyplace. The earlier V3.1 and you can R1 designs continue to be popular and you can discover-pounds, as well as the old deepseek-talk and you can deepseek-reasoner API endpoints were retired to your twenty four July 2026.

DeepSeek Utilize is becoming within the designer preview to own representative utilize developers global. From the modifying the new setup, you need to use the newest OpenAI/Anthropic SDK or programs suitable for the new OpenAI/Anthropic API to gain access to the fresh DeepSeek API. Lookup / Bring API is offered, run on the inside the-home search engine and you can supporting web research and you will get.

Talk assistantRAG / document Q&ACoding agentBulk category Heritage deepseek-talk and you can deepseek-reasoner aliases had been retired; migrate so you can deepseek-v4-thumb. V4-Thumb and V4-Flash-Vision-Exp is resigned; V4-Expert desires approach to V4.1-Flash away from 14 Sep.

DeepSeek patterns and you may what they prices – casino raging bull

After you’ve acquired a keen API secret, you can access the newest DeepSeek model with the following analogy scripts in the OpenAI API style. Distilled versions complement a single 24GB GPU; full models you would like multi-GPU equipment. On the published API costs DeepSeek try much smaller for every million tokens compared to the GPT, Claude and Gemini flagships, and it is competitive to the coding and you may need benchmarks. Our very own cost web page features the full table, cache-strike cost and you will rate history. Discover loads, resources you need, and you can step-by-action settings that have Ollama, vLLM and llama.cpp. See a work, next see just what a comparable frequency will set you back on each provider’s published API prices.

Legacy API aliases retired — surcharge however maybe not effective

Complete desk, rates records as well as the retired V3/R1 endpoints are on the newest DeepSeek costs webpage casino raging bull . Free level, works the current V4 patterns, produced by DeepSeek. I hold the design checklist, API cost and you will sign on actions seemed against DeepSeek’s very own records and printing the newest comment go out on each web page. For advice using the Anthropic API format, excite make reference to Anthropic API.

Use the certified DeepSeek cam

• Come across a great workload, following see what the same volume will set you back for each provider’s wrote API costs.
• Before running this type of instances, set the new DEEPSEEK_API_Trick environment adjustable in your terminal to the API secret.
• Open loads, resources you want, and you can step-by-step configurations with Ollama, vLLM and you can llama.cpp.
• To your wrote API cost DeepSeek are far less per million tokens compared to the GPT, Claude and you may Gemini flagships, and it is aggressive to your programming and need benchmarks.
• All of our cost page have a complete desk, cache-hit cost and rates history.
• This really is a different publication, maybe not the state solution.

DeepSeek posts unlock weights for the GitHub and you can Hugging Deal with, plus they work with in your neighborhood which have Ollama, vLLM, llama.cpp otherwise LM Facility. That is a separate guide and you will community financing. Group discounts, long-perspective surcharges a lot more than a great provider’s threshold, free levels and corporation sales are not included.

Invoke The newest Chat API​

Its latest fundamental model try DeepSeek-V4.1-Flash (API name deepseek-flash), generally readily available while the 9 September 2026, with indigenous image information produced in. DeepSeek AI is an excellent Chinese AI search company you to definitely generates highest words models. The official site are deepseek.com and also the certified cam is actually chat.deepseek.com. The row links to the provider’s very own rates webpage. Text along with a fresh eyes variant.0.150.6Read DeepSeek-V4-ProFlagship cause and you will programming model, to own harder multiple-action work.0.661.98Read DeepSeek’s very own speculative decoding stack vessels equipment-alert schedulers and you will 2–3× token throughput to the H800s.

Try all of our totally free demo chat

Live changelog from design launches, prices changes and infrastructure transform. DeepSeek is actually a great Chinese AI lab whoever higher vocabulary models is actually oddly cost effective to focus on and you can discover-pounds. That is a non-weight analogy, you might place the brand new weight factor in order to real to find stream reaction.

DeepSeek publishes discover loads to your GitHub and Hugging Face, and so they focus on in your town with Ollama, vLLM, llama.cpp otherwise LM Studio. That is another book and you may area money. Batch discounts, long-framework surcharges more than an excellent provider’s tolerance, 100 percent free sections and you will firm sale are not incorporated.

DeepSeek Harness is inside the developer examine to have broker funnel designers global. By modifying the fresh setup, you can utilize the new OpenAI/Anthropic SDK or software programs compatible with the brand new OpenAI/Anthropic API to gain access to the fresh DeepSeek API. Lookup / Bring API has become available, running on our in the-household internet search engine and you can support net research and you may get.