VerdictPal
Beitreten

Modelle / Mistral

Mistral Large 3

Mistral Large · Release 2025-12-02

Mistrals Flaggschiff-MoE Dezember 2025. 262K-Kontext, Vision-Input, $0,50/$1,50 pro 1M Tokens. 675B gesamt / 41B aktive Parameter.

Europäisches Open-API-Frontier für mehrsprachigen Enterprise-Chat, Dokumenten-QA und Vision+Text-Workflows.

#42 von 51 im AA Index · aktueller Snapshot
16 AA Index · 20 Benchmark-Zeilen · 6 QuellenPosition ist benchmark-spezifisch — kein Cross-Family- oder Cross-Source-Ranking.
Composite48
Kontext262K
Input / 1M$0.50
Output / 1M$1.5
Wissensstichtag2025-06-01
Output-Speed64 t/s
TTFT (Default-API)0.59s
Statussolid
Mistral Le Chat
KnowledgeReasoningMathCodingAgenticLong contextTool use
Knowledge
88
Reasoning
71
Math
38
Coding
92
Agentic
12
Long context
51
Tool use
80

62average across 7 tested families

Benchmark-Platzierungen

Wo Mistral Large 3 auf jeder öffentlichen Benchmark-Quelle mit Zeile steht. Rang zählt jedes Modell mit neuester Zeile im selben Test — kein universeller Qualitätsscore.

BenchmarkFamilyRangScoreQuelleDatum
BFCLTool use#8/ 2080Berkeley Function Calling Leaderboard V42026-04-15
MCP AtlasTool use#8/ 1070MCP Atlas leaderboard2026-06-09
LiveCodeBenchCoding#9/ 1347Artificial Analysis2026-07-21
LiveBenchReasoning#11/ 2163LiveBench leaderboard2026-04-15
LongBench v2Long context#11/ 1951LongBench v2 leaderboard2026-04-01
HumanEvalCoding#12/ 1592Mistral Large 3 model card2025-12-02
MMLUKnowledge#13/ 2288Mistral Large 3 model card2025-12-02
AIMEMath#15/ 1938Artificial Analysis2026-07-21
MMLU-ProKnowledge#18/ 2173MMLU-Pro HuggingFace leaderboard2025-12-02
SWE-bench VerifiedCoding#21/ 2376SWE-bench Verified leaderboard2026-06-09
AA output speedPerformance#44/ 69Artificial Analysis2026-07-21
AA time to first tokenPerformance#55/ 69Artificial Analysis2026-07-21
SciCodeCoding#55/ 6536Artificial Analysis2026-07-21
AA-LCRLong context#57/ 6435Artificial Analysis2026-07-21
IFBenchReasoning#57/ 5736Artificial Analysis2026-07-21
Artificial Analysis Intelligence IndexReasoning#58/ 6716Artificial Analysis2026-07-21
τ³-Bench BankingAgentic#61/ 646Artificial Analysis2026-07-21
GPQA DiamondReasoning#62/ 6971Artificial Analysis GPQA Diamond evaluation2026-06-09
Humanity's Last ExamReasoning#64/ 654Artificial Analysis2026-07-21
Terminal-BenchAgentic#64/ 6912Artificial Analysis2026-07-21
Family-Kontext

Die drei höchstscorierenden kartierten Modelle pro Capability-Family. Wo Mistral Large 3 auftaucht, ist es hervorgehoben.

KnowledgeMistral Large 3 · 88
  1. 1
    GPT 5.5
    91
  2. 2
    Claude Opus 4.8
    91
  3. 3
    Claude Sonnet 4.6
    91
ReasoningMistral Large 3 · 71
  1. 1
    Claude Mythos Preview
    95
  2. 2
    GPT-5.6 Sol
    94
  3. 3
    Kimi K3
    94
MathMistral Large 3 · 38
  1. 1
    OpenAI o3
    99
  2. 2
    OpenAI o1
    97
  3. 3
    Gemini 3 Pro
    96
CodingMistral Large 3 · 92
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticMistral Large 3 · 12
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
Long contextMistral Large 3 · 51
  1. 1
    Kimi K3
    75
  2. 2
    GPT-5.1
    75
  3. 3
    GPT-5.6 Sol
    74
Tool useMistral Large 3 · 80
  1. 1
    Claude Sonnet 4.6
    89
  2. 2
    GPT 5.5
    87
  3. 3
    Claude Opus 4.8
    87
Neueste Belege

Benchmark-Zeilen im öffentlichen Ledger für Mistral Large 3 in den letzten 120 Tagen. Ältere Zeilen stehen in der vollen Tabelle unten.

17 neu
BenchmarkFamilyScoreQuelleTage her
AA-LCRLong context34.67%Artificial Analysis2
AA output speedPerformance64 t/sArtificial Analysis2
AA time to first tokenPerformance0.59sArtificial Analysis2
AIMEMath38%Artificial Analysis2
Artificial Analysis Intelligence IndexReasoning15.9%Artificial Analysis2
Humanity's Last ExamReasoning4.1%Artificial Analysis2
IFBenchReasoning36.19%Artificial Analysis2
LiveCodeBenchCoding46.5%Artificial Analysis2
SciCodeCoding36.2%Artificial Analysis2
τ³-Bench BankingAgentic5.77%Artificial Analysis2
Terminal-BenchAgentic11.99%Artificial Analysis2
SWE-bench VerifiedCoding76.2%SWE-bench Verified leaderboard44
GPQA DiamondReasoning71.2%Artificial Analysis GPQA Diamond evaluation44
MCP AtlasTool use70.4%MCP Atlas leaderboard44
BFCLTool use80.1%Berkeley Function Calling Leaderboard V499
LiveBenchReasoning62.8%LiveBench leaderboard99
LongBench v2Long context50.8%LongBench v2 leaderboard113
Family-Abdeckung

Ein hoher Composite, der eine schwache Family versteckt, ist eine Falle. Diese Balken zeigen Families ohne öffentlichen Test — und wo das Modell führt.

Knowledge88
Reasoning71
Math38
Coding92
Agentic12
Long context51
Tool use80
Preis vs. Performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover or tab any dot for name, score, and input price.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMiniMaxxAIMetaZhipuXiaomiMeituanCohere

Volles Benchmark-Ledger

Jeder Katalog-Benchmark für Mistral Large 3. Scores verlinken zur Originalquelle; Lücken heißen: noch keine öffentliche Zeile.

20 Quell-Zeilen

20 of 32 catalog benchmarks have a sourced row for Mistral Large 3.

Knowledge

Reasoning

Math

Coding

Agentic

Long context

Tool use

Performance

Quellen