Meta: Llama 4 Maverick

Llama 4 Maverick 17B Instruct (128E) is a high-capacity multimodal language model from Meta, built on a mixture-of-experts (MoE) architecture with 128 experts and 17 billion active parameters per for…

meta-llama/llama-4-maverick-17b-128e-instruct

Llama 4 Maverick 17B Instruct (128E) is a high-capacity multimodal language model from Meta, built on a mixture-of-experts (MoE) architecture with 128 experts and 17 billion active parameters per forward pass (400B total). It supports multilingual text and image input, and produces multilingual text and code output across 12 supported languages. Optimized for vision-language tasks, Maverick is instruction-tuned for assistant-like behavior, image reasoning, and general-purpose multimodal interaction.

Maverick features early fusion for native multimodality and a 1 million token context window. It was trained on a curated mixture of public, licensed, and Meta-platform data, covering ~22 trillion tokens, with a knowledge cutoff in August 2024. Released on April 5, 2025 under the Llama 4 Community License, Maverick is suited for research and commercial applications requiring advanced multimodal understanding and high model throughput.

Specifications & capabilities

Input
text, image
Output
text
Context window · tokens
524288
Knowledge cutoff
2024-08-31
Published model ID
Llama-4-Maverick-17B-128E-Instruct
Capabilities
multimodal, promptCaching, structuredOutputs, toolCalling, visionInput
Description source ↗

Evaluations

21–40 / 102

OpenRouter Benchmarks

64.478%

source id
openRouterBenchmarks
leaderboard name
GPQA Diamond · Accuracy · Novita
dimension
knowledge
metric
GPQA Diamond · Accuracy
rank
48
sample size
396
aggregation scope
status
published
harness
url
https://github.com/OpenRouterTeam/benchmark-harness
name
OpenRouter benchmark harness
direction
higherIsBetter
run_count
2
conditions
window
newest 90 days; successful visible nonzero runs
dataset
benchmark-leaderboard
benchmark
id
gpqa_diamond
name
GPQA Diamond
record_kind
aggregate
provider_name
Novita
attribution
Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
last_run_at
2026-08-29T17:03:33.297Z
license_url
https://creativecommons.org/licenses/by/4.0/
methodology
Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
metric_kind
GPQA Diamond · Accuracy
rank_method
competition rank among reported evaluations in identical conditions, including unlinked observations
sample_size
396
source_file
benchmark-leaderboard.json
source_page
https://openrouter.ai/benchmarks/gpqa-diamond
measurements
accuracy
0.644781
accuracy_std_dev
0
cost_per_task_usd
0.002468472222222222
output_tokens_per_task
2662.2853535353534
latency_seconds_per_task
55.30653787878788
published_at
2026-11-16T01:46:02.653Z
snapshot_date
2026-11-16
source_revision
7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
transformations
TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
metric_population
59
benchmark_evaluation
true
openrouter_permaslug
meta-llama/llama-4-maverick-17b-128e-instruct
source model id
meta-llama/llama-4-maverick-17b-128e-instruct
Original result ↗

OpenCompass LLM

15.2%

source id
openCompass
leaderboard name
AIME2025
dimension
general
metric
AIME2025
rank
48
aggregation scope
status
published
direction
higherIsBetter
conditions
suite
OverallTable
catalog
open
snapshot
month
REALTIME
fileName
assets/research-rank/research-data.REALTIME.20260309.json
updateTime
2026-03-09
columnFileName
assets/research-rank/research-column.REALTIME.20260309.json
metric_kind
AIME2025
rank_method
competition rank among reported evaluations in identical conditions, including unlinked observations
source_file
0-data.json
source_revision
117c5effa3066abf39f3e8f61109707b09f0cd95223246c07acd6289f4f7fcd2
metric_population
48
benchmark_evaluation
true
source model id
Llama4-Maverick-17B-128E-Instruct
Original result ↗

OpenRouter Benchmarks

64.731%

source id
openRouterBenchmarks
leaderboard name
GPQA Diamond · Accuracy · DeepInfra
dimension
knowledge
metric
GPQA Diamond · Accuracy
rank
54
sample size
396
aggregation scope
status
published
harness
url
https://github.com/OpenRouterTeam/benchmark-harness
name
OpenRouter benchmark harness
direction
higherIsBetter
run_count
2
conditions
window
newest 90 days; successful visible nonzero runs
dataset
benchmark-leaderboard
benchmark
id
gpqa_diamond
name
GPQA Diamond
record_kind
aggregate
provider_name
DeepInfra
attribution
Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
last_run_at
2026-08-29T16:11:23.388Z
license_url
https://creativecommons.org/licenses/by/4.0/
methodology
Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
metric_kind
GPQA Diamond · Accuracy
rank_method
competition rank among reported evaluations in identical conditions, including unlinked observations
sample_size
396
source_file
benchmark-leaderboard.json
source_page
https://openrouter.ai/benchmarks/gpqa-diamond
measurements
accuracy
0.6473065
accuracy_std_dev
0.012626499999999985
cost_per_task_usd
0.002362989898989899
output_tokens_per_task
2763.429292929293
latency_seconds_per_task
57.56465656565656
published_at
2026-11-16T01:46:02.653Z
snapshot_date
2026-11-16
source_revision
7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
transformations
TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
metric_population
71
benchmark_evaluation
true
openrouter_permaslug
meta-llama/llama-4-maverick-17b-128e-instruct
source model id
meta-llama/llama-4-maverick-17b-128e-instruct
Original result ↗

LMArena

1,098.018

source id
lmarena
leaderboard name
LMArena Vision Style Control Humor
dimension
multimodal
metric
Arena rating
rank
101
sample size
214
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1057.3534218701782
confidence upper
1138.6820276082515
Original result ↗

LMArena

1,100.265

source id
lmarena
leaderboard name
LMArena Vision Humor
dimension
multimodal
metric
Arena rating
rank
102
sample size
214
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1059.472468150686
confidence upper
1141.058217331656
Original result ↗

LMArena

1,112.127

source id
lmarena
leaderboard name
LMArena Vision Style Control Creative Writing Vision
dimension
multimodal
metric
Arena rating
rank
104
sample size
214
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1069.5044382285996
confidence upper
1154.7499325682045
Original result ↗

LMArena

1,099.51

source id
lmarena
leaderboard name
LMArena Vision Creative Writing Vision
dimension
multimodal
metric
Arena rating
rank
106
sample size
214
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1056.753253181717
confidence upper
1142.2668328360846
Original result ↗

OpenRouter Benchmarks

66.33%

source id
openRouterBenchmarks
leaderboard name
GPQA Diamond · Accuracy · auto-routing
dimension
knowledge
metric
GPQA Diamond · Accuracy
rank
106
sample size
396
aggregation scope
status
published
harness
url
https://github.com/OpenRouterTeam/benchmark-harness
name
OpenRouter benchmark harness
direction
higherIsBetter
run_count
2
conditions
window
newest 90 days; successful visible nonzero runs
dataset
benchmark-leaderboard
benchmark
id
gpqa_diamond
name
GPQA Diamond
record_kind
aggregate
provider_name
auto-routing
attribution
Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
last_run_at
2026-08-29T19:49:56.943Z
license_url
https://creativecommons.org/licenses/by/4.0/
methodology
Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
metric_kind
GPQA Diamond · Accuracy
rank_method
competition rank among reported evaluations in identical conditions, including unlinked observations
sample_size
396
source_file
benchmark-leaderboard.json
source_page
https://openrouter.ai/benchmarks/gpqa-diamond
measurements
accuracy
0.6632995
accuracy_std_dev
0.005050500000000013
cost_per_task_usd
0.0029102499999999996
output_tokens_per_task
2988.2853535353534
latency_seconds_per_task
81.35334090909092
published_at
2026-11-16T01:46:02.653Z
snapshot_date
2026-11-16
source_revision
7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
transformations
TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
metric_population
132
benchmark_evaluation
true
openrouter_permaslug
meta-llama/llama-4-maverick-17b-128e-instruct
source model id
meta-llama/llama-4-maverick-17b-128e-instruct
Original result ↗

LMArena

1,140.58

source id
lmarena
leaderboard name
LMArena Vision Chinese
dimension
multimodal
metric
Arena rating
rank
108
sample size
294
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1106.5028006907917
confidence upper
1174.6579853037827
Original result ↗

LMArena

1,165.099

source id
lmarena
leaderboard name
LMArena Vision Homework
dimension
multimodal
metric
Arena rating
rank
109
sample size
527
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1139.9356827088757
confidence upper
1190.262080405416
Original result ↗

LMArena

1,147.255

source id
lmarena
leaderboard name
LMArena Vision Style Control Chinese
dimension
multimodal
metric
Arena rating
rank
110
sample size
294
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1112.7051215932693
confidence upper
1181.805008973377
Original result ↗

LMArena

1,166.063

source id
lmarena
leaderboard name
LMArena Vision Style Control Homework
dimension
multimodal
metric
Arena rating
rank
112
sample size
527
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1140.7759063020858
confidence upper
1191.3506476869356
Original result ↗

LMArena

1,153.523

source id
lmarena
leaderboard name
LMArena Vision Style Control Ocr
dimension
multimodal
metric
Arena rating
rank
115
sample size
3079
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1141.6716883842826
confidence upper
1165.373813437432
Original result ↗

LMArena

1,151.809

source id
lmarena
leaderboard name
LMArena Vision Style Control Diagram
dimension
multimodal
metric
Arena rating
rank
116
sample size
946
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1132.3244744606611
confidence upper
1171.2932443431964
Original result ↗

LMArena

1,142.91

source id
lmarena
leaderboard name
LMArena Vision Ocr
dimension
multimodal
metric
Arena rating
rank
117
sample size
3079
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1131.1174900525498
confidence upper
1154.7034490571252
Original result ↗

OpenRouter Benchmarks

42.333%

source id
openRouterBenchmarks
leaderboard name
tau2-bench Verified Airline · Accuracy · auto-routing
dimension
agentic
metric
tau2-bench Verified Airline · Accuracy
rank
117
sample size
100
aggregation scope
status
published
harness
url
https://github.com/OpenRouterTeam/benchmark-harness
name
OpenRouter benchmark harness
direction
higherIsBetter
run_count
2
conditions
window
newest 90 days; successful visible nonzero runs
dataset
benchmark-leaderboard
benchmark
id
tau_bench_verified_airline
name
tau2-bench Verified Airline
record_kind
aggregate
provider_name
auto-routing
attribution
Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter.
last_run_at
2026-08-29T16:33:29.885Z
license_url
https://creativecommons.org/licenses/by/4.0/
methodology
Rows aggregate the benchmark-runs dataset exactly as the leaderboard pages do, over each benchmark's non-zero-score runs from the newest 90 days: accuracy is the task-weighted mean of per-run accuracy, accuracy_std_dev is the population standard deviation of per-run accuracy (null with a single run), and the per-task averages divide summed cost, generation time, and output tokens by summed tasks. Rows are ordered by accuracy descending, then model name.
metric_kind
tau2-bench Verified Airline · Accuracy
rank_method
competition rank among reported evaluations in identical conditions, including unlinked observations
sample_size
100
source_file
benchmark-leaderboard.json
source_page
https://openrouter.ai/benchmarks/tau2-bench-airline
measurements
accuracy
0.4233335
accuracy_std_dev
0.0166665
cost_per_task_usd
0.07010642
output_tokens_per_task
2643.24
latency_seconds_per_task
253.21868
published_at
2026-11-16T01:46:02.653Z
snapshot_date
2026-11-16
source_revision
7ccaf17664252a3053dcb7df8ad0a46e72954a3b4524ddc1c8a7ef8a41b15568
transformations
TensorTrace preserves source results, converts latency to seconds and derives ranks within matching conditions. Individual runs have no derived rank.
metric_population
128
benchmark_evaluation
true
openrouter_permaslug
meta-llama/llama-4-maverick-17b-128e-instruct
source model id
meta-llama/llama-4-maverick-17b-128e-instruct
Original result ↗

LMArena

1,132.931

source id
lmarena
leaderboard name
LMArena Vision Diagram
dimension
multimodal
metric
Arena rating
rank
118
sample size
946
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1113.5955520313642
confidence upper
1152.2658828540045
Original result ↗

LMArena

1,145.404

source id
lmarena
leaderboard name
LMArena Vision Style Control Overall
dimension
multimodal
metric
Arena rating
rank
122
sample size
6912
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1136.2775574100729
confidence upper
1154.5297666773295
Original result ↗

LMArena

1,141.517

source id
lmarena
leaderboard name
LMArena Vision Overall
dimension
multimodal
metric
Arena rating
rank
122
sample size
6912
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1132.4652894428316
confidence upper
1150.569687665915
Original result ↗

LMArena

1,149.727

source id
lmarena
leaderboard name
LMArena Vision English
dimension
multimodal
metric
Arena rating
rank
123
sample size
3378
evaluated at
2026-11-09T00:00:00.000Z
source model id
Llama 4 Maverick 17B 128E Instruct
confidence lower
1137.9872710093707
confidence upper
1161.4670320222572
Original result ↗

Published prices

1–3 / 3

Provider specifications & access

1–2 / 2

Llama-4-Maverick-17B-128E-Instruct ↗

source model id
publishers/hf-meta-llama/models/llama-4-maverick-17b-128e-instruct@001
hosting
vertex-self-deployed
platforms
  • Google Cloud Vertex AI
architecture
input_modalities
  • text
output_modalities
  • text
capabilities

    Llama-4-Maverick-17B-128E-Instruct ↗

    source model id
    publishers/meta/models/llama4@llama-4-maverick-17b-128e-instruct
    hosting
    vertex-self-deployed
    platforms
    • Google Cloud Vertex AI
    architecture
    input_modalities
    • text
    output_modalities
    • text
    capabilities

      Safety assessments

      1–1 / 1

      Foundation Model Transparency Index 2025 ↗

      31 / 100

      source id
      stanfordFMTI
      canonical model id
      meta-llama/llama-4-maverick-17b-128e-instruct
      taxonomy
      transparency
      hazard category
      transparency
      hazard name
      Organization transparency
      policy area
      Meta
      assessment type
      organization-level disclosure
      metric
      disclosure_score
      result label
      31 / 100
      sample size
      100
      updated at
      2026-11-09T18:08:05.227Z

      Source specification observations

      1–3 / 3

      openRouter ↗

      field
      context_length
      value
      524288
      normalized value
      524288
      scope
      source-model
      source id
      openRouter
      confidence
      1

      openRouter ↗

      field
      knowledge_cutoff
      reported field
      knowledge_cutoff
      value
      2024-08-31
      normalized value
      2024-08-31
      precision
      day
      scope
      source-model
      source id
      openRouter
      confidence
      1

      openRouter ↗

      field
      max_output_tokens
      value
      32768
      normalized value
      32768
      scope
      source-model
      source id
      openRouter
      confidence
      1

      Original sources

      1–14 / 14

      ai.meta.com · Model reference ↗

      Hugging Face · meta-llama/Llama-4-Maverick-17B-128E-Instruct ↗

      Google Cloud · Model page ↗

      Google Cloud · Model page ↗

      OpenRouter · Model page ↗

      Epoch AI · Catalog ↗

      GitHub · Catalog ↗

      GitHub · Catalog ↗

      Google Cloud · Catalog ↗

      Hugging Face · Catalog ↗

      LMArena · Catalog ↗

      OpenRouter · Catalog ↗

      OpenRouter · Catalog ↗

      livebench.ai · Catalog ↗

      Source links

      1–11 / 11

      Explore more models

      Models

      Loading
      EVIDENCE / COMPARISON

      Benchmarks

      Explore every benchmark

      Loading

      These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.

      Install TensorTrace