ministral-14b-2512

ministral-14b-2512

The largest model in the Ministral 3 family, Ministral 3 14B is a powerful and efficient language model with vision capabilities.
2025-12-17
LLM
Model capability: imageModel capability: function_call
Input:
$0.33/1M tokens
Output:
$0.33/1M tokens
Bulk order? Contact your manager for exclusive deals

API Overview

Ministral-3-14B-2512 is a high-performance multimodal instruction-tuned model released by Mistral AI. Its core positioning is as the “king of 14B-level reasoning,” delivering state-of-the-art multilingual understanding, image analysis, and complex task reasoning capabilities in edge and on-premises environments.

  • Native Multimodal Architecture: The 14B model directly supports joint text-and-image inputs without the need for external visual modules, making it easy to handle mixed content such as screenshots, charts, and documents.
  • Leapfrogging Reasoning Capabilities: The reasoning variant achieved an accuracy rate of 85% in the AIME ‘25 benchmark, setting a new standard for 14B open-source models.
  • Ultimate Performance-to-Model Size Ratio: Compared to similar models, it achieves higher task completion rates with fewer generated tokens, significantly reducing latency and costs.
  • Deep Support for Over 40 Languages: Specifically optimized for non-English and non-Chinese languages, delivering natural and fluent performance in multilingual conversations, translation, and content creation.
  • Fully Open-Source and Commercially Usable: Offers three versions—base, instruct, and reasoning—all licensed under Apache 2.0 and jointly optimized by NVIDIA and vLLM.

───────────────────────────────────────────────────────────────────

Core Capabilities

👁️ Deep Visual Semantic Parsing: Not only can it recognize image content but also understand interface layouts, data charts, and the logical relationships between text and images.

🧠 High-Level Chain-of-Thought Construction: Demonstrates logic that closely matches large models in tasks such as mathematical proofs, code generation, and multi-hop reasoning.

🌍 Globalized Language Intelligence: Produces outputs that align with local cultural contexts, avoiding mechanical translations and truly enabling natural cross-language interactions.

🛠️ Seamless On-Premise to Cloud Deployment: From DGX Spark and RTX workstations to Jetson edge devices, a single model covers all scenarios for efficient operation.

Playground

Log in to explore more features! Click to Log In

API Analytics

API Reference (1)

API DescriptionAPI EndpointRequest MethodStabilityParameter Description
Chat(Mistral)
POST
Stable
View Details

API Pricing

$
ModelDescriptionContextOfficial Price302.AI Price

ministral-14b-2512

-
256000

Input$0.3 / 1M tokens
Output$0.3 / 1M tokens

Input$0.33/ 1M tokens
Output$0.33/ 1M tokens
10%