Find out how to Entry Ministral 3 fashions with an API

Find out how to Entry Ministral 3 through API

TL;DR

Ministral 3 is a household of open-weight, reasoning-optimized fashions out there in each 3B and 14B variants. The fashions help multimodal reasoning, native operate and gear calling, and an enormous 256K token context window, all launched beneath an Apache 2.0 license.

You possibly can run Ministral 3 straight on Clarifai utilizing the Playground for interactive testing or combine it into your functions by Clarifai’s OpenAI-compatible API.

This information explains the Ministral 3 structure, the best way to entry it by Clarifai, and the way to decide on the appropriate variant on your manufacturing workloads.

Introduction

Trendy AI functions more and more rely upon fashions that may cause reliably, keep lengthy context, and combine cleanly into present instruments and APIs. Whereas closed-source fashions have traditionally led in these capabilities, open-source options are quickly closing the hole. 

Amongst globally out there open fashions, Ministral 3 ranks alongside DeepSeek and the GPT OSS household on the prime tier. Slightly than focusing on leaderboard efficiency on benchmarks, Ministral prioritises performances that matter in manufacturing, akin to producing structured outputs, processing massive paperwork, and executing operate calls inside reside methods.

This makes Ministral 3 well-suited for the calls for of actual enterprise functions, as organisations are more and more adopting open-weight fashions for his or her transparency, deployment flexibility, and skill to run throughout various infrastructure setups, from cloud platforms to on-premise methods.

Ministral 3 Structure

Ministral 3 is a household of dense, edge-optimised multimodal fashions designed for environment friendly reasoning, long-context processing, and native or personal deployment. The household presently consists of 3B and 14B parameter fashions, every out there in base, instruct, and reasoning variants.

Ministral 3 14B

The biggest mannequin within the Ministral household is a dense, reasoning-post-trained structure optimised for math, coding, STEM, and different multi-step reasoning duties. It combines a ~13.5B-parameter language mannequin with a ~0.4B-parameter imaginative and prescient encoder, enabling native textual content and picture understanding. The 14B reasoning variant achieves 85% accuracy on AIME ’25, delivering state-of-the-art efficiency inside its weight class whereas remaining deployable on life like {hardware}. It helps context home windows of as much as 256k tokens, making it appropriate for lengthy paperwork and complicated reasoning workflows.

Ministral 3 3B

The 3B mannequin is a compact, reasoning-post-trained variant designed for extremely environment friendly deployment. It pairs a ~3.4B-parameter language mannequin with a ~0.4B-parameter imaginative and prescient encoder (~4B whole parameters), offering multimodal capabilities. Just like the 14B mannequin, it helps 256k-token context lengths, enabling long-context reasoning and doc evaluation on constrained {hardware}.

Key Technical Options

• Multimodal Capabilities: All Ministral 3 fashions use a hybrid language-and-vision structure, permitting them to course of textual content and pictures concurrently for duties akin to doc understanding and visible reasoning.
• Lengthy-Context Reasoning: Reasoning variants help as much as 256k tokens, enabling prolonged conversations, massive doc ingestion, and multi-step analytical workflows.
• Environment friendly Inference: The fashions are optimised for edge and personal deployments. The 14B mannequin runs in BF16 on ~32 GB VRAM, whereas the 3B mannequin runs in BF16 on ~16 GB VRAM, with quantised variations requiring considerably much less reminiscence.
• Agentic Workflows: Ministral 3 is designed to work nicely with structured outputs, operate calling, and tool-use, making it appropriate for automation and agent-based methods.
• License: All Ministral 3 variants are launched beneath the Apache 2.0 license, enabling unrestricted business use, fine-tuning, and customisation.

Pretraining Benchmark Efficiency

Ministral 3 14B demonstrates robust reasoning capabilities and multilingual efficiency in comparison with equally sized open fashions, whereas sustaining aggressive outcomes on basic data duties. It significantly excels in reasoning-heavy benchmarks and reveals stable factual recall and multilingual understanding.

Accessing Ministral 3 through Clarifai

Stipulations

Earlier than runing  Ministral 3 with the Clarifai API, you’ll want to finish just a few fundamental setup steps:

1. Clarifai Account: Create a Clarifai account to entry hosted AI fashions and APIs.
2. Private Entry Token (PAT): All API requests require a Private Entry Token. You possibly can generate or copy one from the Settings > Secrets and techniques part of your Clarifai dashboard.
   

For added SDKs and setup steering, discuss with the Clarifai Quickstart documentation.

Utilizing the API

The examples beneath use Ministral-3-14B-Reasoning-2512, the biggest mannequin within the Ministral 3 household. It’s optimised for multi-step reasoning, mathematical downside fixing, and long-context workloads, making it well-suited for long-document useecases and agentic functions. Right here’s the best way to make your first API name to the mannequin utilizing completely different strategies.

Python (OpenAI-Suitable)

Python (Clarifai SDK)

You too can use the Clarifai Python SDK for inference with extra management over technology settings. Right here’s the best way to make a prediction and generate streaming output utilizing the SDK:

Node.js (Clarifai SDK)

Right here’s the best way to carry out inference with the Node.js SDK:

Playground

The Clarifai Playground permits you to shortly experiment with prompts, structured outputs, reasoning workflows, and performance calling with out writing any code.

Go to the Playground and select both:

• Ministral-3-3B-Reasoning‑2512
  

• Ministral-3-14B-Reasoning‑2512

Purposes and Use Circumstances

Ministral 3 is designed for groups constructing clever methods that require robust reasoning, long-context understanding, and dependable structured outputs. It performs nicely throughout agentic, technical, multimodal, and business-critical workflows.

Agentic Utility 

Ministral 3 is nicely suited to AI brokers that must plan, cause, and act throughout a number of steps. It may possibly orchestrate instruments and APIs utilizing structured JSON outputs, which makes it dependable for automation pipelines the place consistency issues. 

Lengthy Context

Ministral 3 can analyze massive paperwork utilizing its prolonged 256K token context, making it efficient for summarization, info extraction, and query answering over lengthy technical texts. 

Multimodal Reasoning

Ministral 3 helps multimodal reasoning, permitting functions to mix textual content and visible inputs in a single workflow. This makes it helpful for image-based queries, doc understanding, or assistants that must cause over blended inputs.

Conclusion

Ministral 3 offers reasoning-optimized, open-weight fashions which can be prepared for manufacturing use. With a 256K token context window, multimodal inputs, native device calling, and OpenAI-compatible API entry by Clarifai, it affords a sensible basis for constructing superior AI methods.

The 3B variant is right for low-latency, cost-sensitive deployments, whereas the 14B variant helps deeper analytical workflows. Mixed with Apache 2.0 licensing, Ministral 3 offers groups flexibility, efficiency, and long-term management.

To get began, discover the fashions within the Clarifai Playground or combine them straight into your functions utilizing the API.