Tips, endpoints, model IDs, quick caching and you will code examples — as well as what altered if the dated cam and you will reasoner endpoints was retired. Register from the talk.deepseek.com — the official speak has a free of charge tier and operates the current V4 patterns. The newest mixed enter in speed mixes for every provider’s cached and you will uncached input rates at the cache strike rates; business as opposed to a printed cached-input speed is actually billed from the full type in speed. Newest DeepSeek models and you will API prices for each million tokensModelBest forInput (cache skip)OutputGuideDeepSeek-V4.1-FlashFast casual model. The new organized solution procedure prompts for the DeepSeek’s infrastructure within the Asia, numerous governing bodies limit it to your official products, and you may thinking-holding the brand new open loads prevents giving research anywhere. The earlier V3.step 1 and you will R1 designs continue to be popular and you can unlock-weight, and also the old deepseek-speak and you will deepseek-reasoner API endpoints were resigned to the twenty-four July 2026.
DeepSeek Utilize is within the designer examine for agent funnel developers global. By modifying the brand new setting, you can use the new OpenAI/Anthropic SDK or programs compatible with the fresh OpenAI/Anthropic API to gain access to the newest DeepSeek API. Lookup / Get API is becoming offered, run on the within the-family google and you may supporting online lookup and you will get.
Chat assistantRAG / document Q&ACoding agentBulk classification Legacy deepseek-cam and you can deepseek-reasoner aliases had been retired; migrate so you can deepseek-v4-flash. V4-Flash and you will V4-Flash-Vision-Exp is resigned; V4-Pro requests path to V4.1-Thumb out of 14 Sep.
Casino quick hit: DeepSeek patterns and you can what they rates
Once you have received a keen API key, you have access to the brand new DeepSeek model by using the after the analogy texts in the OpenAI API style. Distilled versions fit a single 24GB GPU; complete models you would like multiple-GPU methods. To your authored API rates DeepSeek is actually far lesser for every million tokens than the GPT, Claude and you can Gemini flagships, and is also aggressive to the coding and you may reason benchmarks. Our cost web page provides an entire dining table, cache-hit cost and rates history. Open loads, methods you need, and you may action-by-step settings with Ollama, vLLM and you can llama.cpp. See a great workload, next see just what an identical frequency will set you back for each provider’s published API rates.
Heritage API aliases retired — surcharge nonetheless maybe not productive
Complete dining table, rate history and the retired V3/R1 endpoints take the new DeepSeek cost page. Free tier, runs the present day V4 patterns, produced by DeepSeek. I secure the model list, casino quick hit API costs and you will log on procedures appeared facing DeepSeek’s own files and you can print the fresh remark time on every webpage. To have examples with the Anthropic API format, please consider Anthropic API.
Use the authoritative DeepSeek talk
- Find a work, up coming see just what an identical volume will cost you on each provider’s published API costs.
- Before running these types of examples, place the fresh DEEPSEEK_API_Trick environment changeable in your critical on the API key.
- Discover loads, equipment you desire, and action-by-step options having Ollama, vLLM and you may llama.cpp.
- For the wrote API rates DeepSeek is actually far smaller for every million tokens compared to the GPT, Claude and you can Gemini flagships, and is aggressive for the coding and you will cause standards.
- The rates page has a full table, cache-hit cost and you will speed records.
- This can be another publication, perhaps not the official provider.
DeepSeek publishes unlock weights on the GitHub and you will Hugging Deal with, and they work with locally with Ollama, vLLM, llama.cpp otherwise LM Facility. That is another book and you may people financing. Batch deals, long-perspective surcharges over a good provider’s threshold, totally free tiers and you will firm sale are not integrated.
Invoke The newest Talk API
The most recent chief model are DeepSeek-V4.1-Flash (API term deepseek-flash), generally available while the 9 Sep 2026, having indigenous photo information built in. DeepSeek AI is actually a Chinese AI research team one makes large code patterns. The official web site is deepseek.com and the formal speak are cam.deepseek.com. All the line website links to your provider’s own price web page. Text message and an experimental sight variation.0.150.6Read DeepSeek-V4-ProFlagship reason and you may coding model, to have harder multiple-action functions.0.661.98Read DeepSeek’s individual speculative decryption pile ships equipment-aware schedulers and you can 2–3× token throughput on the H800s.
Is actually all of our 100 percent free trial speak
Alive changelog out of design launches, costs changes and you may structure transform. DeepSeek is a good Chinese AI research whose high language designs are surprisingly inexpensive to work at and unlock-weight. That is a non-stream example, you can set the brand new load parameter in order to correct to locate load response.
DeepSeek publishes open loads to the GitHub and you will Hugging Face, and they work with locally with Ollama, vLLM, llama.cpp or LM Business. That is another guide and you may community financing. Batch discounts, long-context surcharges more than an excellent provider’s threshold, free levels and you can corporation product sales commonly included.
DeepSeek Utilize has become inside designer preview for agent utilize builders international. Because of the altering the brand new arrangement, you can utilize the brand new OpenAI/Anthropic SDK or software programs suitable for the newest OpenAI/Anthropic API to gain access to the new DeepSeek API. Look / Fetch API is becoming readily available, run on our inside the-house search and you can help net look and you may bring.