BenchAtlas

Rankings / Mistral AI

Mistral Medium 3

released 2025-05-07

BenchAtlas Index

as of 2026-07-14
22.8
base configuration
rank #399
11 families · 6 categories · high

Benchmark evidence

16 results
agentic coding
τ²-Bench24.3 %independentArtificial Analysis
coding
LiveCodeBench v6
implementation=artificial-analysis
40.0 %independentArtificial Analysis
SciCode33.1 %independentArtificial Analysis
external indices
Intelligence Index v4.112.5 pointsindependentArtificial Analysis
AA Math Index30.3 pointsindependentArtificial Analysis
factuality
MASK42.6 %
39.345.9
independentScale Labs
knowledge science
GPQA Diamond
implementation=artificial-analysis
57.8 %independentArtificial Analysis
Humanity's Last Exam
contamination=Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. · implementation=scale
4.5 %
3.75.3
independentScale Labs
Humanity's Last Exam
implementation=artificial-analysis
4.3 %independentArtificial Analysis
MMLU-Pro
implementation=artificial-analysis
76.0 %independentArtificial Analysis
long context instruction
AA-LCR28.0 %independentArtificial Analysis
IFBench39.3 %independentArtificial Analysis
multimodal
VISTA34.6 %
33.535.7
independentScale Labs
reasoning math
AIME
implementation=artificial-analysis
44.0 %independentArtificial Analysis
AIME
year=2025 · implementation=artificial-analysis
30.3 %independentArtificial Analysis
MATH-500
implementation=artificial-analysis
90.7 %independentArtificial Analysis

Agent + model results

systems, not bare-model scores
agent + model Artificial Analysis harness + Mistral Medium 3Terminal-Bench Hard3.8 %independentArtificial Analysis

These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.

Related variants