Ministral 3 3B Instruct 2512

Compact 3B multimodal model for cost-sensitive assistants, tools, and lightweight reasoning.

About model

Ministral 3 3B Instruct is a compact 3B-class multimodal workhorse that combines a 3.4B language backbone with a 0.4B vision encoder. It preserves the 256K token context window and instruction-following behavior of larger Ministral 3 variants while targeting low-latency, cost-sensitive workloads. Ideal for routing, extraction, simple assistants, and high-volume automation pipelines where throughput and price matter more than frontier-level reasoning.

Parameters

3.8B

3.4B language model plus 0.4B vision encoder for lightweight multimodal tasks.

Context Window

256K

Shared long-context behavior with larger Ministral models for consistent prompting patterns.

Multimodal IO

Vision

Handles basic image and document understanding alongside text for utility-style assistants and tools.

Model key capabilities

Lightweight Multimodal Core: Handles text and basic image tasks without the overhead of larger frontier models
High-Volume Automation: Optimized for routing, extraction, tagging, and short-form replies in cost-sensitive pipelines
Multilingual & Code-Aware: Supports many languages and code tasks for classification, transformation, and analysis
Family Compatibility: Shares APIs and prompting patterns with Ministral 3 8B and 14B so you can swap models without friction

Quickstart guides

RAG

Building a RAG Workflow

Agents

Agent Workflows

Apps

Next.js Chat Quickstart

Performance benchmarks

Model	AIME 2025	GPQA Diamond	HLE	LiveCodeBench	MATH500	SWE-bench verified
Ministral 3 3B Instruct 2512	72.1%	53.4%		54.8%
Related open-source models
Competitor closed-source models
Claude Opus 4.6		90.5%	34.2%			78.7%
OpenAI o3		83.3%	24.9%		99.2%	62.3%
OpenAI o1		76.8%			96.4%	48.9%
GPT-4o		49.2%	2.7%	32.3%	89.3%	31.0%

Model card
Architecture overview:
• 3.4B parameter language backbone paired with a 0.4B vision encoder for unified multimodal IO.
• 256K token context window aligned with the rest of the Ministral 3 lineup for consistent long-context behavior.
• Instruction-tuned objective tailored for concise, schema-following outputs suitable for automation and routing.
Training and performance:
• Trained on multilingual and code-heavy corpora to keep quality competitive despite the small parameter budget.
• Emphasis on robustness and stability for narrow, repetitive tasks rather than open-ended frontier reasoning.
• Strong cost-per-token characteristics, making it attractive for high-QPS backends and batch workloads.
Applications & use cases
Automation and decisioning:
• High-volume classification, tagging, routing, and triage of tickets, events, or user messages.
• Information extraction from short documents, forms, and logs into structured records.
• Policy enforcement or guardrail-style checks that filter, normalize, or annotate content.
User-facing assistants and utilities:
• Lightweight chatbots embedded in products or workflows where fast responses matter more than deep reasoning.
• Multimodal utilities that inspect screenshots or small images for quick explanation, labeling, or checks.
• Localized, task-specific helpers (FAQ bots, small copilots, inline explainers) that can run at very low cost.

Related models

Model specifications

Model data

Model provider
Mistral AI
Type
Chat
Main use cases
Chat
Small & Fast
Deployment
On-Demand Dedicated
Parameters
3.8B
Context length
256K
Input modalities
Text
Output modalities
Text

Released
October 31, 2025
Quantization level
FP8
External link
Provider docs
Category
Chat

Quickstart docs

Deploy model

Ministral 3 3B Instruct 2512

About model

Model card

Applications & use cases