AB
AB
AiBoss

All Categories

project

DeepSeek V3.1 - The latest open-source AI model version of DeepSeek

DeepSeek V3.1 is the latest AI model version released by DeepSeek, featuring a hybrid inference architecture that allows for seamless switching between thinking and non-thinking modes, significantly improving thinking efficiency. The model has undergone several upgrades compared to V3, including...

What is DeepSeek V3.1?

DeepSeek V3.1 is DeepSeek's latest AI model version, featuring a hybrid inference architecture that allows for seamless switching between thinking and non-thinking modes, significantly improving thinking efficiency. The model has undergone several upgrades compared to V3, with the context window expanded from 64k to 128k, enabling it to handle longer text inputs. It uses a hybrid expert (MoE) architecture with the same number of parameters as V3. The model excels in programming and search agents, efficiently completing complex tasks. The model is now updated across DeepSeek's web platform, app, mini-program, and API open platform, and is self-identified as DeepSeek V3.

Main features of DeepSeek V3.1

  • Natural Language ProcessingDeepSeek V3.1 can generate high-quality text, such as creative writing and story creation, and provides more lively and informative answers to questions.
  • Programming skillsThe model can generate more complex and more complete code.
  • Multidisciplinary knowledgeIt provides more accurate and informative answers to niche historical questions, and also offers in-depth solutions in the fields of science and technology.
  • Hybrid Inference ArchitectureIt supports free switching between thinking mode and non-thinking mode to adapt to different scenario needs.
  • Highly efficient thinking abilityCompared to its predecessor, it offers higher thinking efficiency, more concise output, and comparable performance.
  • Powerful Agent CapabilitiesIt excels in programming and search agent tasks, and can perform complex code repair and multi-step inference search.
  • API Functionality Upgrade:support deepseek-chat and deepseek-reasoner Two modes, context extended to 128K, supporting Anthropic API format.

Project address for DeepSeek V3.1

  • HuggingFace model library:
    • Base Modelhttps://huggingface.co/deepseek-ai/DeepSeek-V3.1-Base
    • Post-trained modelhttps://huggingface.co/deepseek-ai/DeepSeek-V3.1

How to use DeepSeek V3.1

  • Web versionVisit the DeepSeek website and use the latest version of the model.
  • AppDownload and install the DeepSeek App, log in to your account, and use the V3.1 model.
  • Mini ProgramSearch for the DeepSeek mini-program on WeChat or other platforms, log in to your account, and then use it.
  • API Open PlatformTo use the API, you need to register and obtain an API key first, and then follow the guidelines in the official documentation. The API supports a thinking mode (…).deepseek-reasoner) and non-thinking mode (deepseek-chatThe context window expands to 128K and supports...strictPattern-based Function Calling.
  • Access the Hugging Face platformVisit the Hugging Face DeepSeek-V3.1-Base model page: https://huggingface.co/deepseek-ai/DeepSeek-V3.1-Base
    • Model DownloadDownload the model weights, configuration files, and script code.
    • Local deploymentInstall the necessary dependencies (such as Python, Transformers library, etc.) in the local environment, load the model and perform inference.
    • API callsUse the API provided by Hugging Face to integrate DeepSeek-V3.1 models into your own applications or services to achieve automated processing and interaction.

DeepSeek V3.1 API Pricing

The API call price will be adjusted starting at midnight on September 6, 2025, and the nighttime discount will be canceled.

  • Enter price:
    • Cache hit: 0.5 yuan/million tokens
    • Cache miss: 4 yuan/million tokens
  • Output Price12 yuan/million tokens

Application Scenarios of DeepSeek V3.1

  • Content creationIt is used to generate creative texts such as articles, stories, and poems, helping creators to inspire themselves and improve their creative efficiency.
  • Programming aidsIt helps developers quickly generate code frameworks, optimize code logic, and improve programming efficiency, making it suitable for front-end development and mini-game development.
  • EducationAs a teaching aid, it provides students with answers to knowledge questions and explains complex scientific and historical issues.
  • Scientific researchIt assists researchers in organizing and analyzing data, and provides answers and analytical approaches to scientific questions.
Back