Inference
EN
Mistral

Mistral Large 4

Mistral Large 4 is a reasoning model from Mistral. 40 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-10
RankedEnough results for an index.
IndexAll categories, out of 100.60.7±4.1Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.113/sMedian 67/s
First tokenSeconds to the first output token.19.4 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.70Median $0.60
Output / 1MUS dollars per 1M output tokens.$2.10Median $2.55
ContextMaximum tokens in one request.1MMedian 262K
EloLMArena rating and rank.1414±9 #136Median 1422

Run locally

1000B parameters
16-bit
2101.0 GB
8-bit
1116.6 GB
4-bit
637.6 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.15 counted66.340 / 218
CodingCode writing and repair.8 counted62.755 / 209
ReasoningLogic problems and puzzles.7 counted59.269 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.4 counted43.7173 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted57.7—
MathMath problems.5 counted61.644 / 168

Results

40 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis CyberGym-E2E-AAAgenticAbove fitFar above this model's index.79.4%2 / 19
OTIS Mock AIME 2024-2025Math98.9% ±0.422 / 200
GPQA diamond · Epoch AIKnowledge89.1% ±1.850 / 217
Artificial Analysis Cyber IndexAgenticAbove fitFar above this model's index.50.9%5 / 19
LiveBench MathematicsMath93.6%18 / 66
GDPval-AA normalizedAgentic46.1%37 / 133
Vibe Code Bench v1.1Coding78.4% ±3.535 / 123
BrowseCompAgentic84.5%18 / 63
Critical Physics TasksReasoning10.0%65 / 214
LiveBench Agentic CodingAgentic57.2%22 / 66
Artificial Analysis CWE-Bench-AAAgentic58.3%7 / 19
LiveBench Data AnalysisReasoning76.5%33 / 80
Terminal-Bench 4.0Agentic27.8% ±1.826 / 61
FrontierMath-Tiers-1-3-v2-PrivateMath58.6% ±2.945 / 90
Agent Arena command recoveryAgentic1.825 / 50
Code MigrationCoding30.6% ±4.241 / 77
LiveBench ReasoningReasoning83.9%43 / 80
FrontierMath-Tier-4-v2-PrivateMath26.8% ±7.039 / 72
Chess PuzzlesReasoning31.0% ±4.750 / 90
LiveBench CodingCoding77.2%37 / 66
Mystery Game PuzzlesReasoning18.5% ±4.048 / 80
Artificial Analysis Omniscience AccuracyKnowledge26.0%99 / 164
LiveBench Instruction FollowingInstruction64.8%42 / 66
EBR-benchReasoning18.1% ±3.221 / 31
IOICoding45.3% ±3.831 / 45
Artificial Analysis AutomationBenchAgentic57.8%16 / 23
Artificial Analysis DeepsecBench-AAAgentic15.0%15 / 21
Agent Arena task outcomeAgentic−4.739 / 54
CWE-bench v2Agentic45.6%8 / 11
Agent Arena steerabilityAgentic−6.441 / 54
Artificial Analysis Terminal-Bench v4.0Agentic24.7%17 / 22
Medical Long Context Reasoning (MLCR-AA)Reasoning17.8%17 / 22
FrontierCode 2.0Coding19.7%21 / 27
MirrorCodeCoding27.8% ±8.931 / 39
Vibe Code Bench 1-100Coding14.3% ±3.221 / 26
LiveBench DesignAgentic65.6%43 / 52
SimpleQA VerifiedKnowledge21.2% ±1.377 / 91
Artificial Analysis GDP.pdfAgentic17.6%19 / 22
Furniture AssemblyReasoning25.8% ±6.032 / 37
Multimodal Knowledge Agent BenchmarkAgentic35.1%17 / 19
LiveBench Coding 2.0Coding60.0%47 / 52
ProofBench v1.1Math10.0% ±3.049 / 54
LiveBench Browser UseAgentic8.0%45 / 49
LiveBench Long AgenticAgentic11.8%48 / 51
LiveBench LanguageKnowledge49.6%65 / 66
LiveBench ResearchAgentic28.1%52 / 52
HLAB-Vals-1.1Legal17.5% ±3.28 / 35
Finance Agent (v2)Finance54.7% ±0.626 / 79
Tax Agent BenchFinance30.4% ±3.333 / 82
MedScribeHealthcare80.4% ±2.057 / 109
Legal Research BenchLegal31.7% ±3.251 / 90
MedCodeHealthcare40.7% ±2.261 / 106
EMBFinance56.2% ±3.247 / 75
BioMysteryBenchScience67.0% ±2.720 / 30
MysteryMechanismScience12.2% ±2.230 / 30
b3: Backbone Breaker BenchmarkAgentic97.1%—
Refusal of harmful cyber requests (JailbreakBench, StrongREJECT, AgentHarm)Agentic96.6%—
CybenchAgentic93.0%—
Finch (FinWorkBench): Finance & Accounting across Spreadsheet-Centric Enterprise WorkflowsAgentic73.3%—
STEMBench120: six-domain scientific research progressReasoning2.5%—

Sources

BenchLMCC BY-NC 4.0Epoch AICC BYLiveBenchNo licence statedVals AINo licence statedLMArenaCC BY 4.0

History

5 releases
Index60.2 → 60.7
23 Oct 202612 Nov 2026
Output / 1M$2.09 → $2.10
23 Oct 202612 Nov 2026

Questions

How much does Mistral Large 4 cost?
$0.70 per 1M input tokens, $2.10 per 1M output tokens.
How good is Mistral Large 4?
Its index is 60.7 out of 100. That is place 67 of 283 ranked models.
What is Mistral Large 4 best at?
Agentic: 66.3 out of 100, place 40 of 218.
How fast is Mistral Large 4?
113 output tokens per second, 19.4 s to the first token.
What is the context window of Mistral Large 4?
1M tokens.
Is Mistral Large 4 open source?
Yes. Its weights are published.
When was Mistral Large 4 released?
October 2026.

Badge

Mistral Large 4 on Inference360: place 67 of 283, index 60.7
HTML
<a href="https://inference360.ai/models/mistral-large-4"><img src="https://inference360.ai/badges/mistral-large-4.svg" alt="Mistral Large 4 on Inference360: place 67 of 283, index 60.7" height="20"></a>
Markdown
[![Mistral Large 4 on Inference360: place 67 of 283, index 60.7](https://inference360.ai/badges/mistral-large-4.svg)](https://inference360.ai/models/mistral-large-4)

Compare with

GLM-5.260.6Qwen3.7 Max60.5Gemini 3.1 Pro Preview61.5DeepSeek V4 Flash 073161.9Mistral Medium 3.542.7

Lower price

dots3-note Preview63.4 · FreeGLM-5.3-Flash63.1 · $0.078/1MLing 3.1 Flash67.2 · $0.135/1M

More from Mistral

All models
Leanstral 1.5UnrankedMistral Medium 3.542.7Mistral Medium 3.5 128B50.3Mistral Small 437.9LeanstralUnranked