Inference
EN

Mistral Large 4 vs Mistral Medium 3.5

Change models
Mistral Large 4Mistral
Mistral Medium 3.5Mistral
IndexAll categories, out of 100.
60.7 ±4.1 best
42.7 ±8.9
GapIndex points below the top.
Top
−18.0
CoverageIndex weight that has results.
85% best
65%
SpeedOutput tokens per second.
113/s best
92/s
EloLMArena rating and rank.
1414
1421 best
ReleasedMonth of release.
2026-10
2026-04
Input / 1MUS dollars per 1M input tokens.
$0.70 best
$1.50
Output / 1MUS dollars per 1M output tokens.
$2.10 best
$7.50
Batch output / 1MBatch price per 1M output tokens.
—
$3.75
1K chats2K tokens in, 500 out, each.
$2.45 best
$6.75
1K documents20K tokens in, 1K out, each.
$16.10 best
$37.50
1K agent tasks200K tokens in, 20K out, each.
$182.00 best
$450.00
ContextMaximum tokens in one request.
1M best
262K
AgenticMulti-step tasks with tools.
66.3 best
47.1
CodingCode writing and repair.
62.7 best
47.1
ReasoningLogic problems and puzzles.
59.2
—
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
43.7 best
24.3
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
57.7
—
MathMath problems.
61.6 best
50.3

Benchmarks

12 shared
Only Mistral Large 4: 48Only Mistral Medium 3.5: 11
Mistral Large 4
Mistral Medium 3.5
Vibe Code Bench v1.1Coding
78.4% best35 / 123
2.9%116 / 123
MedScribeHealthcare
80.4% best57 / 109
67.7%99 / 109
MedCodeHealthcare
40.7% best61 / 106
33.8%83 / 106
Legal Research BenchLegal
31.7% best51 / 90
9.1%82 / 90
Tax Agent BenchFinance
30.4% best33 / 82
4.5%79 / 82
Finance Agent (v2)Finance
54.7% best26 / 79
32.1%69 / 79
Code MigrationCoding
30.6% best41 / 77
5.1%71 / 77
EMBFinance
56.2% best47 / 75
11.2%73 / 75
Agent Arena steerabilityAgentic
−6.4 best41 / 54
−16.950 / 54
Agent Arena task outcomeAgentic
−4.7 best39 / 54
−16.849 / 54
ProofBench v1.1Math
10.0% best49 / 54
9.0%50 / 54
Agent Arena command recoveryAgentic
1.8 best25 / 50
−20.447 / 50