VerdictPal
Join

Models / Meta

Llama 4 Maverick

Llama 4 · Released 2026-04-05

Meta's reasoning-focused Llama 4 MoE. 1M context, 17B active / 400B total params, 128 experts. Open weights for multimodal assistants and coding.

Default Llama 4 pick for quality-first self-hosting when Scout's 10M context isn't required.

#154 of 168 on AA Index · current snapshot
15 AA Index · 21 benchmark rows · 6 sourcesPosition is benchmark-specific — not a cross-family or cross-source ranking.
Composite49
Context1M
Input / 1M$0.17
Output / 1M$0.60
Knowledge cutoff2024-08-01
Output speed90 t/s
TTFT (default API)0.54s
Statusflagship

Family profile

Best published score in each covered benchmark family.

66/100 avg
Knowledge
86
Reasoning
69
Math
89
Coding
92
Agentic
8
Multimodal
61
Long context
50
Tool use
76

8 tested benchmark families

Benchmark placements

Where Llama 4 Maverick places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

BenchmarkFamilyRankScoreSourceDate
MMMUMultimodal#6/ 861MMMU leaderboard2026-04-05
MATHMath#10/ 2589Artificial Analysis2026-09-01
LongBench v2Long context#12/ 1950LongBench v2 leaderboard2026-04-01
LiveBenchReasoning#13/ 2162LiveBench leaderboard2026-04-15
HumanEvalCoding#14/ 1592Meta Llama 4 model card2026-04-05
BFCLTool use#15/ 2076Berkeley Function Calling Leaderboard V42026-04-15
MMLUKnowledge#19/ 2286Meta Llama 4 model card2026-04-05
SWE-bench VerifiedCoding#22/ 2375SWE-bench Verified leaderboard2026-06-09
LiveCodeBenchCoding#28/ 3140Artificial Analysis2026-09-01
MMLU-ProKnowledge#29/ 3981Artificial Analysis2026-09-01
AIMEMath#34/ 3719Artificial Analysis2026-09-01
AA output speedPerformance#45/ 187Artificial Analysis2026-09-01
AA time to first tokenPerformance#78/ 186Artificial Analysis2026-09-01
τ³-Bench BankingAgentic#104/ 1064Artificial Analysis2026-09-01
IFBenchReasoning#116/ 12743Artificial Analysis2026-09-01
AA-LCRLong context#150/ 17950Artificial Analysis2026-09-01
SciCodeCoding#159/ 18033Artificial Analysis2026-09-01
GPQA DiamondReasoning#166/ 18569Artificial Analysis GPQA Diamond evaluation2026-06-09
Humanity's Last ExamReasoning#173/ 1805Artificial Analysis2026-09-01
Artificial Analysis Intelligence IndexReasoning#174/ 18715Artificial Analysis2026-09-01
Terminal-BenchAgentic#177/ 1848Artificial Analysis2026-09-01
Family context

The three highest-scoring models with pages in each capability family. Where Llama 4 Maverick shows up, it's highlighted.

KnowledgeLlama 4 Maverick · 86
  1. 1
    GPT 5.5
    91
  2. 2
    Claude Opus 4.8
    91
  3. 3
    Claude Sonnet 4.6
    91
ReasoningLlama 4 Maverick · 69
  1. 1
    Claude Mythos Preview
    95
  2. 2
    Gemini 3.7 Flash
    95
  3. 3
    Grok 4.6
    95
MathLlama 4 Maverick · 89
  1. 1
    OpenAI o3
    99
  2. 2
    GPT-5 (medium)
    99
  3. 3
    GPT-5 (high)
    99
CodingLlama 4 Maverick · 92
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticLlama 4 Maverick · 8
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
Long contextLlama 4 Maverick · 50
  1. 1
    Muse Spark 1.2
    83
  2. 2
    Kimi K3
    83
  3. 3
    Gemini 3.5 Flash
    81
Tool useLlama 4 Maverick · 76
  1. 1
    Claude Sonnet 4.6
    89
  2. 2
    GPT 5.5
    87
  3. 3
    Claude Opus 4.8
    87
Newest receipts

Benchmark rows added to the public ledger for Llama 4 Maverick in the last 120 days. Older rows live in the full table below.

15 recent
BenchmarkFamilyScoreSourceDays ago
AA-LCRLong context50%Artificial Analysis5
AA output speedPerformance90 t/sArtificial Analysis5
AA time to first tokenPerformance0.54sArtificial Analysis5
AIMEMath19.33%Artificial Analysis5
Artificial Analysis Intelligence IndexReasoning14.5%Artificial Analysis5
Humanity's Last ExamReasoning4.9%Artificial Analysis5
IFBenchReasoning42.99%Artificial Analysis5
LiveCodeBenchCoding39.7%Artificial Analysis5
MATHMath88.87%Artificial Analysis5
MMLU-ProKnowledge80.9%Artificial Analysis5
SciCodeCoding33.1%Artificial Analysis5
τ³-Bench BankingAgentic3.71%Artificial Analysis5
Terminal-BenchAgentic7.87%Artificial Analysis5
SWE-bench VerifiedCoding74.8%SWE-bench Verified leaderboard89
GPQA DiamondReasoning69.4%Artificial Analysis GPQA Diamond evaluation89
Family coverage

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

Knowledge86
Reasoning69
Math89
Coding92
Agentic8
Multimodal61
Long context50
Tool use76
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover any dot for name, score, and input price. Keyboard: tab through the top twelve, or use the ranking below.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMetaMiniMaxxAIZhipuXiaomiOtherMeituanCohereNVIDIA

Full benchmark ledger

Every catalog benchmark for Llama 4 Maverick. Scores link to the original source; gaps mean no public row exists yet.

21 sourced rows

21 of 37 catalog benchmarks have a sourced row for Llama 4 Maverick.

Knowledge

Reasoning

Math

Coding

Agentic

Multimodal

Long context

Tool use

Performance

Sources