Ministral 3 14B 2512
Mistral's efficiency model. Context window: 262K tokens.
Scores by test
Methodology →What you need to know
Ministral 3 14B 2512 is a low-cost model optimized for persona consistency and structured data handling. With a blended cost of $0.20 per million tokens, it provides a high value-to-price ratio for developers needing reliable output formats and tool calling without the expense of frontier models. Its 262K context window is substantial for its size, and internal testing confirms it maintains performance across long-context tasks.
The model performs consistently across most technical categories, scoring 4/5 in structured output, tool calling, and strategic analysis. However, it shows a significant deficiency in safety calibration, scoring 1/5, meaning it lacks robust internal guardrails. It also struggles with complex agentic planning compared to its other capabilities.
Use this model for high-volume tasks requiring strict persona adherence, structured JSON outputs, or long-document processing on a tight budget. Skip this model if your application requires autonomous multi-step planning or strict safety filtering for end-user interactions.
Strengths — Top 3
Relative weaknesses — Bottom 3
Similar models