Mistral Large and Mistral Small Release

Mistral AI has announced the release of Mistral Large, its most advanced language model to date, and Mistral Small, a model optimized for cost and latency. Mistral Large is positioned as a flagship model capable of complex multilingual reasoning, text understanding, and code generation, ranking as the second-ranked model generally available through an API, trailing only GPT-4.

Mistral Large Technical Capabilities

Mistral Large is designed for high-reasoning tasks and provides several key technical enhancements over previous models:

  • Multilingual Fluency: The model is natively fluent in English, French, Spanish, German, and Italian, incorporating a nuanced understanding of cultural context and grammar.
  • Context Window: It features a 32K tokens context window, enabling precise information recall from large documents.
  • Instruction Following: The model's precise instruction-following capabilities allow developers to implement custom moderation policies; Mistral AI used this feature to establish the system-level moderation for le Chat.
  • Function Calling: Mistral Large is natively capable of function calling, which, combined with a constrained output mode, supports large-scale application development and tech stack modernization.

Performance Benchmarks

Mistral Large demonstrates strong performance across reasoning, knowledge, and coding benchmarks:

  • General Reasoning: In MMLU (Measuring massive multitask language understanding) benchmarks, Mistral Large (pre-trained) is compared alongside GPT-4, Claude 2, Gemini Pro 1.0, GPT 3.5, and LLaMA 2 70B.
  • Multilingual Performance: The model strongly outperforms LLaMA 2 70B on HellaSwag, Arc Challenge, and MMLU benchmarks specifically in French, German, Spanish, and Italian.
  • Math and Coding: Mistral Large shows top-tier performance on popular benchmarks including HumanEval pass@1, MBPP pass@1, Math maj@4, and GSM8K (maj@8 and maj@1).

Mistral Small and Endpoint Updates

Alongside the flagship model, Mistral AI introduced Mistral Small, an intermediary solution between open-weight models and Mistral Large. Mistral Small is optimized for low-latency workloads and outperforms Mixtral 8x7B.

Mistral Small inherits the same RAG-enablement and function calling innovations as Mistral Large. The updated endpoint offering now includes:

  • Optimized Endpoints: mistral-small-2402 and mistral-large-2402.
  • Open-weight Endpoints: open-mistral-7B and open-mixtral-8x7b.
  • Legacy Endpoints: mistral-medium remains available but is not being updated.

Developer Tools: JSON Mode and Function Calling

To improve integration into production pipelines, Mistral AI has introduced JSON format mode and function calling. JSON format mode forces the model to output valid JSON, allowing for structured data extraction. Function calling allows developers to interface Mistral endpoints with internal APIs, databases, or code.

Currently, these features are only available on the mistral-small and mistral-large models.

Deployment and Availability

Mistral AI models are available through three primary channels:

  1. la Plateforme: Hosted on Mistral's European infrastructure.
  2. Azure: Mistral Large is available via Azure AI Studio and Azure Machine Learning.
  3. Self-deployment: Model weights are available for sensitive use cases requiring deployment in a private environment.

Sources

Related

  • Dispatch
    Mistral Large 2 release notes / what's new

    Mistral AI has released Mistral Large 2, a 123B parameter model designed for high throughput single-node inference with competitive performance in coding, reasoning, and multilingual capabilities.

  • Dispatch
    Mistral Small 3.1 Release Notes

    Mistral AI has released Mistral Small 3.1, an Apache 2.0 licensed model featuring multimodal understanding, a 128k token context window, and performance that surpasses GPT-4o Mini and Gemma 3.

  • Dispatch
    Pixtral Large Release Notes

    Mistral AI has released Pixtral Large, a 124B open-weights multimodal model that achieves state-of-the-art performance on MathVista, DocVQA, and VQAv2 while maintaining the text capabilities of Mistral Large 2.

  • Dispatch
    Mistral Small 3 Release Notes

    Mistral AI has released Mistral Small 3, a latency-optimized 24B-parameter model under the Apache 2.0 license that rivals the performance of models three times its size.

  • Dispatch
    Mistral Small 4 release notes / what's new

    Mistral AI has released Mistral Small 4, a unified open-source model under Apache 2.0 that integrates reasoning, multimodal, and agentic coding capabilities into a single versatile architecture.