Inference
EN

Mistral Medium 3 vs Mistral Small 3.1

Change models
Mistral Medium 3Mistral
Mistral Small 3.1Mistral
IndexAll categories, out of 100.
27.1 ±5.2 best
24.2 ±4.1
GapIndex points below the top.
Top
−2.9 · no clear gap
CoverageIndex weight that has results.
95%
95%
SpeedOutput tokens per second.
94/s best
28/s
EloLMArena rating and rank.
1370
—
ReleasedMonth of release.
2025-05
2025-03
Input / 1MUS dollars per 1M input tokens.
$0.40
—
Output / 1MUS dollars per 1M output tokens.
$2.00
—
1K chats2K tokens in, 500 out, each.
$1.80
—
1K documents20K tokens in, 1K out, each.
$10.00
—
1K agent tasks200K tokens in, 20K out, each.
$120.00
—
ContextMaximum tokens in one request.
131K best
128K
AgenticMulti-step tasks with tools.
18.9
26.7 best
CodingCode writing and repair.
9.3
25.6 best
ReasoningLogic problems and puzzles.
32.8 best
29.7
MultimodalTasks with images and text.
24.1
28.8 best
KnowledgeFacts and expert knowledge.
31.1 best
25.7
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
25.7 best
17.0
MathMath problems.
32.2 best
16.7

Benchmarks

10 shared
Only Mistral Medium 3: 4Only Mistral Small 3.1: 21
Mistral Medium 3
Mistral Small 3.1
Artificial Analysis Humanity's Last ExamKnowledge
4.1%199 / 222
4.3% best190 / 222
Artificial Analysis Long Context ReasoningReasoning
31.3% best179 / 217
22.3%191 / 217
GPQA diamond · Epoch AIKnowledge
59.5% best129 / 217
44.7%164 / 217
Critical Physics TasksReasoning
0.0%150 / 214
0.0%150 / 214
OTIS Mock AIME 2024-2025Math
32.2% best131 / 200
4.9%170 / 200
Artificial Analysis GPQA DiamondKnowledge
57.8% best161 / 199
45.4%182 / 199
Artificial Analysis Omniscience AccuracyKnowledge
18.3% best139 / 164
15.1%152 / 164
τ²-Bench Tool-Agent-User EvaluationAgentic
24.3%124 / 149
25.1% best120 / 149
Artificial Analysis IFBenchInstruction
39.3% best107 / 146
29.9%135 / 146
MATH level 5Math
81.6% best27 / 88
46.8%51 / 88