這是用戶在 2024-10-15 12:13 為 https://mistral.ai/technology/#models 保存的雙語快照頁面,由 沉浸式翻譯 提供雙語支持。了解如何保存?

Mistral technology

AI models AI 模型

We release the world’s most capable open models, enabling frontier AI innovation.
我們發布世界上最強大的開放模型,推動前沿 AI 創新。

Developer platform 開發者平台

Our portable developer platform serves our open and optimized models for building fast and intelligent applications. Get started for free!
我們的便攜開發者平台提供開放且優化的模型,用於構建快速且智能的應用程式。立即免費開始!

Premier models 高級模型

State of the art models across a variety of sizes, available to experiment under the Mistral Research License and take to production with the Commercial License.
各種尺寸的尖端模型,可在 Mistral 研究授權下進行實驗,並在商業授權下投入生產。

Mistral Small 24.09

Enterprise-grade small model.
企業級小型模型。

  • The most powerful model in its size
    同類型中最強大的模型
  • Available under the Mistral Research License
    根據 Mistral 研究許可條款提供
  • 128k token context window
    128k token 上下文窗口
  • Cost-efficient and fast model for a wide array of use cases such as translation, summarization, and sentiment analysis
    適用於翻譯、摘要和情感分析等多種用途的高效且快速的模型
Mistral Large 2

Top-tier reasoning for high-complexity tasks, for your most sophisticated needs.
針對高複雜度任務的頂尖推理,滿足您最高級的需求。

  • Multi-lingual (incl. European languages, Chinese, Japanese, Korean, Hindi, Arabic)
    多語言(包括歐洲語言、中文、日文、韓文、印地語、阿拉伯語)
  • Large context window of 128K tokens
    大型上下文窗口,包含 128K 個標記
  • Native function calling capacities and JSON outputs
    原生函數呼叫能力和 JSON 輸出
  • High coding proficiency (80+ coding languages)
    高編程能力(80+種編程語言)
Codestral

State-of-the-art Mistral model trained specifically for code tasks.
專為程式碼任務訓練的最新 Mistral 模型。

  • Trained on 80+ programming languages (incl. Python, Java, C, C++, PHP, Bash)
    訓練於 80+種編程語言(包括 Python、Java、C、C++、PHP、Bash)
  • Optimized for low latency: Way smaller than competitive coding models
    優化低延遲:比競爭對手的編程模型小得多
  • Context window of 32K tokens
    32K 標記的上下文視窗
Mistral Embed

State-of-the-art semantic for extracting representation of text extracts.
最先進的語義技術,用於提取文本片段的表示。

  • English only for now 目前僅限英文
  • Achieves a retrieval score of 55.26 on the Massive Text Embedding Benchmark (MTEB)
    在大規模文本嵌入基準測試(MTEB)上達到 55.26 的檢索分數

For more details on the various pricing options, check out our pricing page here: See pricing.
有關各種定價選項的更多詳細信息,請參閱我們的定價頁面:查看定價。

Free models 免費模型

Free to use under the Apache 2.0 license.
根據 Apache 2.0 許可證免費使用。

Latest models  最新模型
Pixtral 12B

Version-capable small model.
支持版本的小型模型。

  • Analyze, search, review, and better understand images
    分析、搜索、審查和更好地理解圖像
  • Available under the Apache 2.0 license
    根據 Apache 2.0 授權條款提供
  • Deploy in your own environment so you don’t have to upload your files to a third party provider
    在自己的環境中部署,這樣您就不需要將文件上傳到第三方提供商
Mathstral

Variant of Mistral-7B, optimized for solving advanced mathematics problems.
Mistral-7B 的變體,針對解決高級數學問題進行了優化。

  • Context window of 32K tokens
    32K 標記的上下文視窗
  • 7B parameters 7B 參數
  • Available under Apache 2.0 license
    根據 Apache 2.0 授權條款提供
Codestral Mamba

A Mamba2 language model designed for coding tasks.
專為編程任務設計的 Mamba2 語言模型。

  • Context window of 256K tokens
    256K 標記的上下文視窗
  • 7.3B parameters 7.3B 參數
  • Available under Apache 2.0 license
    根據 Apache 2.0 授權條款提供
Mistral NeMo

A state-of-the-art 12B small model built in collaboration with NVIDIA.
與 NVIDIA 合作開發的最新 12B 小型模型。

  • The most powerful model in its size category
    同類型中最強大的模型
  • Available under Apache 2.0 license
    根據 Apache 2.0 授權條款提供
  • Multi-lingual (incl. European languages, Chinese, Japanese, Korean, Hindi, Arabic)
    多語言(包括歐洲語言、中文、日文、韓文、印地語、阿拉伯語)
  • Large context window of 128K tokens
    大型上下文窗口,包含 128K 個標記
Mistral 8 X 22B

Mixtral 8x22B set a new standard for performance and efficiency, with only 39B active parameters out of 141B, offering unparalleled cost efficiency for its size. It is natively capable of function calling, which enables application development and tech stack modernisation at scale.
Mixtral 8x22B 設定了新的性能和效率標準,僅使用 39B 的活動參數,總共 141B,提供了其規模的無與倫比的成本效益。它原生支持函數調用,這使得應用程序開發和技術堆棧現代化成為可能。

    Mistral 8 X 7B

    A high-quality sparse mixture of experts (SMoE) with open weights. Matches or outperforms GPT3.5 on most standard benchmarks, particularly in multilingual capabilities and code.
    一個高品質的稀疏專家混合(SMoE),開放權重。在大多數標準基準測試中,特別是在多語言能力和代碼方面,匹配或超越 GPT3.5。

      Mistral 7B

      The first Mistral model, engineered for superior performance and efficiency. The model leverages grouped-query attention (GQA) for faster inference, coupled with sliding window attention (SWA) to effectively handle sequences of arbitrary length with a reduced inference cost.
      第一個 Mistral 模型,設計用於提供卓越的性能和效率。該模型利用分組查詢注意力(GQA)以加快推理速度,並結合滑動窗口注意力(SWA)來有效處理任意長度的序列,從而降低推理成本。

        These models also have commercial licenses for business purposes: Explore commercial licenses.
        這些模型也有商業許可證供商業用途:探索商業許可證。

        Fine-tune our models 微調我們的模型

        We allow you to fine-tune our models in an easy, effective & cost-efficient way, and thus use smaller and better-suited models to solve your specific use cases. Fine-tuning can be done with our open-source fine-tuning code as well as on La Plateforme with our efficient Fine-tuning API.
        我們讓您能夠以簡單、有效且成本效益高的方式微調我們的模型,從而使用較小且更適合的模型來解決您的特定用例。微調可以使用我們的開源微調程式碼,也可以在 La Plateforme 上使用我們的高效微調 API 進行。

        Fine-tune Mistral models on La Plateforme with Mistral Fine-tuning API
        使用 Mistral 微調 API 在 La Plateforme 上微調 Mistral 模型

        Leverage Mistral’s unique expertise in training models by using our highly efficient fine-tuning service to specialize both our open-source and commercial models.
        利用 Mistral 在訓練模型方面的獨特專業知識,使用我們高效的微調服務來專門化我們的開源和商業模型。

        Use Mistral fine-tuning code
        使用 Mistral 微調程式碼

        Benefit from Mistral fine-tuning code to perform fine-tuning on Mistral open-source models on your own.
        利用 Mistral 的微調程式碼,自行對 Mistral 的開源模型進行微調。

        Pay-as-you-go pricing 按使用量計費

        Price in $ 價格(美元)
        Price in € 價格(歐元)
        Premier models 高級模型
        ModelAPI Name API 名稱Description 描述Input (/M tokens) 輸入(/M 代幣)Output (/M tokens) 輸出(/M 代幣)
        Mistral Large 2mistral-large-2407Top-tier reasoning for high-complexity tasks, for your most sophisticated needs.
        針對高複雜度任務的頂尖推理,滿足您最高級的需求。
        $2$6
        Mistral Small 24.09mistral-small-2409Cost-efficient, fast, and reliable option for use cases such as translation, summarization, and sentiment analysis.
        成本效益高、快速且可靠的選擇,適用於翻譯、摘要和情感分析等用途。
        $0.2$0.6
        Codestralcodestral-2405State-of-the-art Mistral model trained specifically for code tasks.
        專為程式碼任務訓練的最新 Mistral 模型。
        $0.2$0.6
        Mistral Embedmistral-embedState-of-the-art semantic for extracting representation of text extracts.
        最先進的語義技術,用於提取文本片段的表示。
        $0.1
        ModelAPI NameDescriptionInput (/M tokens)Output (/M tokens)
        Mistral Large 2mistral-large-2407Top-tier reasoning for high-complexity tasks, for your most sophisticated needs.1.8€5.4€
        Mistral Small 24.09mistral-small-2409Cost-efficient, fast, and reliable option for use cases such as translation, summarization, and sentiment analysis.0.18€0.54€
        Codestralcodestral-2405State-of-the-art Mistral model trained specifically for code tasks.0.18€0.54€
        Mistral Embedmistral-embedState-of-the-art semantic for extracting representation of text extracts.0.09€
        Free models 免費模型
        ModelAPI Name API 名稱Description 描述Input (/M tokens) 輸入(/M 代幣)Output (/M tokens) 輸出(/M 代幣)
        Pixtral 12Bpixtral-12bVersion-capable small model.
        支持版本的小型模型。
        $0.15$0.15
        Mistral NeMomistral-nemoState-of-the-art Mistral model trained specifically for code tasks.
        專為程式碼任務訓練的最新 Mistral 模型。
        $0.15$0.15
        Mistral 7Bopen-mistral-7bA 7B transformer model, fast-deployed and easily customisable.
        一個 7B 的變換器模型,快速部署且容易自訂。
        $0.25$0.25
        Mixtral 8x7Bopen-mixtral-8x7bA 7B sparse Mixture-of-Experts (SMoE). Uses 12.9B active parameters out of 45B total.
        一個 7B 稀疏的混合專家模型(SMoE)。使用 45B 總參數中的 12.9B 活躍參數。
        $0.7$0.7
        Mixtral 8x22Bopen-mixtral-8x22bMixtral 8x22B is currently the most performant open model. A 22B sparse Mixture-of-Experts (SMoE). Uses only 39B active parameters out of 141B.
        Mixtral 8x22B 目前是最高效能的開放模型。這是一個 22B 稀疏的混合專家模型(SMoE)。僅使用 141B 中的 39B 活躍參數。
        $2$6
        ModelAPI NameDescriptionInput (/M tokens)Output (/M tokens)
        Pixtral 12Bpixtral-12bVersion-capable small model.0.13€0.13€
        Mistral NeMomistral-nemoState-of-the-art Mistral model trained specifically for code tasks.0.13€0.13€
        Mistral 7Bopen-mistral-7bA 7B transformer model, fast-deployed and easily customisable.0.2€0.2€
        Mixtral 8x7Bopen-mixtral-8x7bA 7B sparse Mixture-of-Experts (SMoE). Uses 12.9B active parameters out of 45B total.0.65€0.65€
        Mixtral 8x22Bopen-mixtral-8x22bMixtral 8x22B is currently the most performant open model. A 22B sparse Mixture-of-Experts (SMoE). Uses only 39B active parameters out of 141B.1.9€5.6€
        Fine-tuning 微調
        ModelOne-off training (/M tokens)
        一次性訓練(/M 代幣)
        Storage 儲存Input (/M tokens) 輸入(/M 代幣)Output (/M tokens) 輸出(/M 代幣)
        Mistral NeMo$1$2 per month per model
        每月每模型 2 美元
        $0.15$0.15
        Mistral Large 2$9$4 per month per model
        每月每模型 4 美元
        $2$6
        Mistral Small$3$2 per month per model
        每月每模型 2 美元
        $0.2$0.6
        Codestral$3$2 per month per model
        每月每模型 2 美元
        $0.2$0.6
        ModelOne-off training (/M tokens)StorageInput (/M tokens)Output (/M tokens)
        Mistral NeMo0.9€1.8€ per month per model0.13€0.13€
        Mistral Large 28.2€3.8€ per month per model1.8€5.4€
        Mistral Small2.7€1.8€ per month per model0.18€0.54€
        Codestral2.7€1.8€ per month per model0.18€0.54€

        Tokens are numerical representations of words or parts of words. On average, one token is roughly equivalent to 4 characters or 0.75 words in English.
        代幣是單詞或部分單詞的數字表示。平均來說,一個代幣大約相當於 4 個字符或 0.75 個英文單詞。

        Fine-tuning pricing explained
        微調價格說明

        Mistral AI provides a fine-tuning API through La Plateforme, making it easy to fine-tune our open-source and commercial models. There are three costs related to fine-tuning:
        Mistral AI 通過 La Plateforme 提供微調 API,使得微調我們的開源和商業模型變得容易。微調相關的費用有三種:

        • One-off training: Price per token on the data you want to fine-tune our standard models on; minimum fee per fine-tuning job of $4
          一次性訓練:根據您想要微調的數據中的每個標記收費;每次微調工作的最低費用為 $4
        • Inference: Price per input/output token when using the fine-tuned model(s)
          推理:使用微調模型時,每個輸入/輸出標記的價格
        • Storage: Price per month per model for storage (irrespective of model usage; models can be deleted any time)
          存儲:模型存儲的每月價格(與模型使用無關;模型可以隨時刪除)

        Deploy anywhere 隨處部署

        La Plateforme
        La Plateforme

        Get started with Mistral models in a few clicks via our developer platform hosted on Mistral’s infrastructure and build your own applications and services. Our servers are hosted in EU.
        幾次點擊即可在我們的開發者平台上開始使用 Mistral 模型,該平台由 Mistral 的基礎設施主辦,您可以建立自己的應用程式和服務。我們的伺服器位於歐盟。

        Cloud platforms
        Cloud platforms 雲平台

        Access our models via your preferred cloud provider and use your cloud credits. Our open models are currently available via our cloud partners (GCP, AWS, Azure, IBM, Snowflake, NVIDIA).
        通過您偏好的雲端提供商訪問我們的模型,並使用您的雲端信用額度。我們的開放模型目前可通過我們的雲端合作夥伴(GCP、AWS、Azure、IBM、Snowflake、NVIDIA)獲取。

        Mistral Large 2 is available on Azure AI Studio, AWS Bedrock, Google Cloud Model Garden, IBM Watsonx, and Snowflake.
        Mistral Large 2 可在 Azure AI Studio、AWS Bedrock、Google Cloud Model Garden、IBM Watsonx 和 Snowflake 上獲取。

        Self-deployment
        Self-deployment 自行部署

        Deploy Mistral models on virtual cloud or on-prem. Self-deployment offers more advanced levels of customisation and control. Your data stays within your walls. Try deploying our open models, and contact our team to deploy our optimized models similarly.
        在虛擬雲端或本地部署 Mistral 模型。自行部署提供更高級的自訂和控制。您的數據將保留在您的範圍內。嘗試部署我們的開放模型,並聯繫我們的團隊以類似方式部署我們的優化模型。

        La Plateforme

        Access our latest products via our developer platform, hosted in Europe
        通過我們的開發者平台訪問我們最新的產品,該平台位於歐洲

        from mistralai.client import MistralClient
        from mistralai.models.chat_completion import ChatMessage
        
        api_key = os.environ["MISTRAL_API_KEY"]
        model = "mistral-tiny"
        
        client = MistralClient(api_key=api_key)
        
        messages = [
            ChatMessage(role="user",
            content="Who is the most renowned French painter?")
        ]
        
        Built for developers 針對開發者設計

        La Plateforme is developers’ preferred way to access all Mistral Al’s models. Hosted and served on Mistral’s infrastructure, in Europe.
        La Plateforme 是開發者最喜愛的訪問所有 Mistral AI 模型的方式。由 Mistral 基礎設施在歐洲托管和提供服務。

        • Our best models at the best price: Get access to our models at an unmatched price/performance point
          最佳模型,最佳價格:以無可匹敵的價格/性能比獲取我們的模型
        • Guides & community: Use our guides and community forums to build your own application and services
          使用我們的指南和社群論壇來建立您自己的應用程式和服務
        • Secure by design: Your data are encrypted at rest (AES256) and in transit (TLS 1.2+); our servers are in the EU
          設計即安全:您的數據在靜止時(AES256)和傳輸時(TLS 1.2+)都是加密的;我們的伺服器位於歐盟
        • Get started for free! No credit card required
          免費開始使用!無需信用卡
        from mistralai.client import MistralClient
        from mistralai.models.chat_completion import ChatMessage
        
        api_key = os.environ["MISTRAL_API_KEY"]
        model = "mistral-tiny"
        
        client = MistralClient(api_key=api_key)
        
        messages = [
            ChatMessage(role="user",
            content="Who is the most renowned French painter?")
        ]
        

        Model licenses 模型授權

        We distribute two categories of models:
        我們分發兩類模型:

        • Models licensed under Apache 2.0 that are no-string-attached contributions to the community and industry
          根據 Apache 2.0 授權的模型,這些模型是對社區和行業無條件的貢獻
        • Models with a double license: Mistral Research License / Mistral Commercial License. These models can be used for non-commercial purposes with no strings attached. A commercial license must be bought for commercial use cases requiring self-deployment. These models may also be used directly through our portable platform or distributors, with pay-as-you-go and reserved offerings.
          雙重授權的模型:Mistral 研究授權 / Mistral 商業授權。這些模型可以用於非商業用途,無需任何附加條件。商業用途需購買商業授權。這些模型也可以通過我們的便攜平台或分銷商直接使用,提供按需付費和預訂服務。

        Apache 2.0Mistral Research License Mistral 研究授權Mistral Commercial License
        Mistral 商業授權
        Access to weights 訪問權重
        Deployment for research purposes and individual usage
        用於研究目的和個人使用的部署
        Creation of derivatives (e.g. fine-tuning) for research purposes and individual usage
        創建衍生品(例如微調)以用於研究目的和個人使用
        The same license applies to derivatives
        相同的授權適用於衍生品
        The same license applies to derivatives
        相同的授權適用於衍生品
        Deployment for commercial purposes (internal & external use cases)
        商業用途的部署(內部及外部使用案例)
        Requires Mistral Commercial License
        需要 Mistral 商業授權
        Creation and usage of derivatives (e.g. fine-tuning) for commercial use cases
        創建和使用衍生物(例如微調)以供商業用途
        Requires Mistral Commercial License
        需要 Mistral 商業授權
        Custom terms & support (self-deployment)
        自訂條款與支援(自行部署)