NOUMENOpen app
NOUMEN

Mistral · Open weights

Ministral 3 14B

Added Dec 2, 2025Open weights

Compare models

Choose up to 3 models to compare.0 of 3 chosen

Loading models…

Use in the app

Ministral 3 14B is a balanced model in the Ministral 3 family, designed for edge deployment. A powerful, efficient language model with vision capabilities, fine-tuned for instruction tasks. Features multilingual support, strong system prompt adherence, and native function calling. Apache 2.0 licensed.

  • Vision
Input
Text, Images
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.10
Output
$0.40

Reference prices as published by the model provider. API example

Specifications

Context window
262.1K
Max output
32.8K
Average output (7 days)
203 tokens
Longer than 22% of 468 models

Side by side

Compare models

Put Ministral 3 14B next to up to three other models. The comparison stays in the link, so you can share it.

Compare models

Choose up to 3 models to compare.0 of 3 chosen

Loading models…

TryHermes 3 70BPhi 4 MultimodalCeleris 1All 3

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

Intelligence Index

6.0

Better than 3% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    14.4

    Composite of the coding evaluations

    Better than 8% of 168 scored models

  • Math Index

    30.0

    Composite of the math evaluations

    Better than 27% of 118 scored models

Agentic work

  • AutomationBench-AA

    0.6%

    Workflow automation, penalized for guardrail violations

    Better than 6% of 173 scored models

  • AutomationBench-AA tasks completed

    0.2%

    Workflows finished end to end without a guardrail violation

    Better than 13% of 173 scored models

  • AA-Briefcasev1.1

    131 Elo

    Agentic knowledge work, rated head to head

    Better than 1% of 166 scored models

  • GDPval-AAv2.1

    247 Elo

    Economically valuable occupational tasks, rated head to head

    Better than 2% of 187 scored models

Document reasoning

  • GDP.pdf

    1.2%

    Professional reasoning over PDF documents, all-pass rate

    Better than 4% of 171 scored models

  • AA-LCRv1.1

    26.3%

    Reasoning across long documents

    Better than 11% of 309 scored models

  • MLCR-AA

    2.2%

    Long-context reasoning over medical records

    Better than 5% of 121 scored models

Reasoning and knowledge

  • Humanity’s Last Exam

    4.6%

    Expert-level questions across academic fields

    Better than 13% of 323 scored models

  • GPQA Diamond

    57.2%

    Graduate-level science questions

    Better than 15% of 291 scored models

  • MMLU-Pro

    69.3%

    Multi-subject knowledge and reasoning

    Better than 19% of 132 scored models

  • IFBench

    32.0%

    Following precise output instructions

    Better than 9% of 215 scored models

Coding

  • Terminal-Benchv4.0

    0.0%

    Practical software engineering in a terminal

    Better than 0% of 171 scored models

  • SciCode

    23.8%

    Scientific computing in Python

    Better than 4% of 174 scored models

  • LiveCodeBench

    35.1%

    Recent competitive programming problems

    Better than 27% of 129 scored models

Math

  • AIME 2025

    30.0%

    Competition mathematics, 2025 problems

    Better than 27% of 118 scored models

Legacy evaluations

  • Terminal-Bench Hard

    4.5%

    Earlier terminal task suite

    Better than 10% of 212 scored models

  • τ²-Bench Telecom

    27.2%

    Tool use in a customer-support simulation

    Better than 18% of 214 scored models

  • AA-LCR (unversioned)

    26.3%

    Long-context reasoning, earlier grading

    Better than 11% of 309 scored models

Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Prompt cache hits
24.9%
Request success (24h)
100.0%
Reliable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 100.0%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistralai/ministral-14b-instruct-2512",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="mistralai/ministral-14b-instruct-2512",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

Mistral

Other Mistral models, starting with the same family.