Atlas

Models

← All models

Ministral 3 14B

Mistral AI · 2025-12-02 · 13.9B parameters

Ministral 3 14B (`ministral-14b-2512`) is Mistral AI's hosted December 2025 instruction-post-trained 14B-class dense multimodal model, optimized for local and edge deployment. It accepts text and image inputs, produces multilingual text, supports a 262,144-token context window, and has Apache-2.0 open weights; the Base and Reasoning models are separate released checkpoints.

Benchmark scores

BenchmarkScore
AA-LCR22.0
AA-Omniscience Index-66.8
AIME 202530.0
Artificial Analysis Agentic Index2.2
Artificial Analysis Coding Index14.4
Artificial Analysis Intelligence Index11.1
Artificial Analysis Omniscience Accuracy12.3
Artificial Analysis Omniscience Hallucination Rate90.2
Artificial Analysis Openness Index38.9
BuseyBench SVG2.4
Capability128.2
CritPt0.0
Design Arena Overall Code Categories1095
FormationEval87.7
GDPval-AA v2484
GDPval-AA v2 (normalized)0.0
Global MMLU Lite68.3
GPQA Diamond57.2
Humanity's Last Exam (Text-Only)4.6
IFBench (Artificial Analysis)32.0
LiveCodeBench (Artificial Analysis)35.1
MMMU Pro49.8
ObviousBench41.0
SciCode (Artificial Analysis)23.6
Terminal-Bench 2.19.7
Terminal-Bench Hard4.5
τ²-bench (Telecom)27.2
𝜏³-Banking6.6
Loading Atlas data…