Mistral AI API

Mistral Small Creative is a specialized Labs model variant optimized for creative content generation. It builds on the Mistral Small architecture with adjustments for more imaginative and varied outputs in writing tasks.

Ministral 14B is part of the Ministral 3 family, a 14B parameter multimodal model with vision capabilities under Apache 2.0. It offers advanced capabilities for local deployment with instruct, base, and reasoning variants achieving 85% on AIME'25.

Devstral 2 is a 123B parameter dense transformer coding model achieving 72.2% on SWE-bench Verified with 256K context. Released under modified MIT license, it's the state-of-the-art open model for code agents, 7x more cost-efficient than Claude Sonnet.

Mistral Medium 3.1 is Mistral's frontier-class multimodal model released August 2025 with 128K context. It delivers near-frontier performance at $0.4/$2 per million tokens, excelling in reasoning, coding, and enterprise workflows.

Mistral Medium 3.1 (August 2025) is a frontier-class multimodal model with improved tone and performance. It features 128K context, native vision, and enhanced reasoning for STEM and enterprise workflows at competitive pricing.

Devstral Medium is a high-performance agentic coding model for complex software engineering tasks, achieving 61.6% on SWE-Bench Verified. It's designed for generalization across prompt styles and tool use in code agents and frameworks.

Devstral Medium is a high-performance agentic coding model achieving 61.6% on SWE-Bench Verified. It excels at complex software engineering tasks across entire codebases, surpassing GPT-4.1 and Gemini 2.5 Pro in code-related tasks at a fraction of the cost.

Devstral Small is a 24B parameter agentic coding model built with All Hands AI, achieving 46.8% on SWE-Bench Verified. Released under Apache 2.0, it can run locally on a single RTX 4090 or 32GB RAM Mac for autonomous software development.

Voxtral Mini is a 3B parameter open-source speech model built on Ministral 3B under Apache 2.0. It handles transcription, Q&A from audio, and multilingual speech understanding for up to 40 minutes of audio, optimized for edge deployment.

Voxtral Small is a 24B parameter speech understanding model built on Mistral Small 3.1 under Apache 2.0. It supports 30-minute transcription, 40-minute audio understanding, Q&A, summarization, and function calling from voice in 8+ languages.

Voxtral Small 24B is an open-source speech understanding model built on Mistral Small 3.1 under Apache 2.0. It handles transcription, translation, Q&A, and summarization directly from audio in 8+ languages with 32K token context.

Mistral Small 3.2 is a 24B parameter multimodal model with 128K context, improved instruction following, and reduced repetition. It handles text and images, runs on single RTX 4090 when quantized, and delivers 150 tokens/second under Apache 2.0.

Mistral Small 3.2 improves on 3.1 with better instruction following (84.78% vs 82.75%), reduced infinite generations (1.29% vs 2.11%), and more robust function calling. It maintains the 24B/128K context architecture under Apache 2.0.

Devstral Small is a 24B parameter agentic LLM for software engineering, achieving 46.8% on SWE-Bench Verified. Released under Apache 2.0, it runs locally on consumer GPUs and excels at solving real-world GitHub issues autonomously.

Mistral Medium 3 delivers frontier performance at $0.4/$2 per million tokens, performing at 90%+ of Claude Sonnet 3.7 across benchmarks. It's deployable on 4+ GPUs and surpasses Llama 4 Maverick and Cohere Command A.

Mistral Small 3.1 is a 24B multimodal model with 128K context, supporting text and image inputs. It outperforms GPT-4o Mini and Gemma 3 while delivering 150 tokens/second, released under Apache 2.0 for commercial use.

Magistral Medium is Mistral's enterprise reasoning model with chain-of-thought capabilities, scoring 73.6% on AIME2024 (90% with majority voting). It excels in multilingual step-by-step reasoning for legal, financial, and scientific applications.

Magistral Small is a 24B parameter open-source reasoning model under Apache 2.0, achieving 70.7% on AIME2024. It provides traceable, multilingual chain-of-thought reasoning in English, French, Spanish, German, Italian, Arabic, Russian, and Chinese.

Mistral Saba is a 24B parameter regional model trained for Arabic and South Asian languages including Tamil and Malayalam. It outperforms models 5x its size on Arabic benchmarks while providing culturally relevant responses.

Mistral Small 3 is a 24B parameter latency-optimized model achieving ~81% MMLU accuracy at 150 tokens/second. It's designed for fast-response conversational agents and low-latency function calling under Apache 2.0.

Mistral Large 2 (24.11) includes improvements in long context understanding, system prompts, and function calling accuracy. Released alongside Pixtral Large, it's optimized for RAG and agentic workflows in enterprise deployments.

Mistral Large 3 is a 675B parameter sparse MoE model (41B active) trained on 3000 H200 GPUs, representing Mistral's frontier open-weight multimodal model. It supports 256K context, native vision, and excels in agentic workflows and enterprise applications.

Pixtral Large is a 124B parameter open-weights multimodal model built on Mistral Large 2, achieving frontier-level image understanding. It processes up to 30 high-resolution images per input with 128K context, excelling in document and chart analysis.

Ministral 3B is a compact 3B parameter model optimized for edge deployment on phones, laptops, and IoT devices. It delivers robust multimodal capabilities in a small footprint, suitable for low-resource environments under Apache 2.0.

Ministral 8B is an 8B parameter model offering best-in-class text and vision capabilities for single-GPU operation. It provides an excellent balance of performance and efficiency for edge deployment and embedded applications.

Ministral 3B is a compact 3B parameter multimodal model from the Ministral 3 family with vision capabilities. It runs on consumer hardware and edge devices, offering text and image understanding with 256K context in a 3-4GB quantized footprint.

Ministral 8B is an 8B parameter multimodal model offering best-in-class text and vision capabilities for edge deployment. It supports single-GPU operation and provides an optimal balance of performance and efficiency under Apache 2.0.

Pixtral 12B is Mistral's first multimodal model with 12B text decoder + 400M vision encoder under Apache 2.0. It processes images at native resolution with 128K context, excelling in document QA and visual reasoning without compromising text performance.

Mistral Large 2 (24.07) is a 123B parameter model with 128K context, significantly upgraded for long context understanding and function calling. It delivers top-tier performance for enterprise use cases including knowledge exploration and automation.

Mistral Nemo is a 12B parameter model developed with NVIDIA featuring 128K context and the Tekken tokenizer. It's state-of-the-art in its class for reasoning, world knowledge, and coding in 11+ languages under Apache 2.0.

Mistral Nemo is a 12B parameter model built with NVIDIA featuring 128K context and the Tekken tokenizer trained on 100+ languages. It excels in multilingual tasks, coding, and reasoning, serving as a drop-in replacement for Mistral 7B.

Codestral is Mistral's cutting-edge code generation model supporting 80+ programming languages with optimized low-latency performance. It specializes in fill-in-the-middle completion, code correction, and test generation with 2.5x faster performance than its predecessor.

Mistral 7B Instruct v0.3 features an extended vocabulary with v3 Tokenizer and function calling support. It enhances language understanding and generation while maintaining the efficient 7B parameter architecture under Apache 2.0.

Mixtral 8x22B is a sparse MoE model with 141B total / 39B active parameters, 64K context, and native function calling. It outperforms Llama 2 70B and matches GPT-3.5 while being cost-efficient under Apache 2.0.

Mistral Large is Mistral's flagship large model for high-complexity enterprise tasks with strong reasoning, knowledge, and coding capabilities. It supports function calling and excels in RAG and agentic workflows across multiple languages.

Mistral 7B Instruct v0.2 introduces a 32K context window and improved performance over v0.1. It outperforms Llama 2 13B and Llama 1 34B on most benchmarks while remaining efficient for local deployment under Apache 2.0.

Mistral Tiny is an earlier lightweight Mistral model optimized for speed and efficiency. It provides basic language capabilities for simple tasks where minimal latency and resource usage are prioritized over maximum performance.

Mixtral 8x7B is a sparse MoE model with 45B total / 13B active parameters using 8 experts per layer. It outperforms Llama 2 70B and GPT-3.5 while running 6x faster, mastering English, French, German, Spanish, and Italian.

Mistral 7B is Mistral's foundational 7.3B parameter open-source model under Apache 2.0, using sliding window attention and grouped-query attention. It outperforms Llama 2 13B on all benchmarks while being efficient enough for consumer hardware.

Mistral 7B Instruct is the instruction-tuned version of Mistral 7B, fine-tuned on publicly available datasets. It outperforms all 7B models on MT-Bench and competes with 13B chat models while maintaining Apache 2.0 licensing.

Mistral 7B Instruct v0.1 is the original instruction-tuned version of Mistral 7B released September 2023. It demonstrates strong instruction-following capabilities while maintaining efficiency through sliding window and grouped-query attention.