Language modelActive

Llama 4 Maverick

Specifications and access

Specification	Value and source
Weights	Open weightsSource
Access	API, LocalSource
Parameters	400BSource
Active parameters	17BSource
Context window	1,048,576 tokensSource
Knowledge cutoff	August 2024Source
Notes	128 experts, 17B active per tokenSource
License	Llama 4 Community LicenseSource
Release	2025-04-05Source
Parameters	400 BSource
Active parameters	17 BSource
Architecture	early-fusion sparse mixture-of-expertsSource
Input	Text, ImageSource
Output	Text, codeSource
Languages	12Source
Training tokens	~22TSource
Knowledge cutoff	2024-08Source

Published benchmarks

MultiNRC

Overall

Native reasoning tasks in French, Spanish, and Chinese that were not translated from English templates. Three languages counter English-only evaluation, but they do not represent global language coverage.

1.GPT-5 Pro65.2%

2.Gemini 3.1 Pro Preview64.74%

3.GPT-5.4 Pro62.27%

4.GPT-5.458.29%

5.GPT-552.13%

6.o3-pro49%

7.Gemini 2.5 Pro45.12%

8.DeepSeek-R127.58%

9.GPT-4.121.23%

10.Llama 4 Maverick8.44%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-multinrc-native-languages
Metric: accuracy
Scale: 0-100
Participants: 10
Methodology and context: Native, non-translated reasoning tasks with automatic evaluation.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 8.44% ± 1.68. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-bc7fb4bc0d9a. Source payload hash: bc7fb4bc0d9ad4757d46043b7c57b0b056967f2b0062249a99a3f0b616b95e23. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

PRBench Finance

Overall

600 realistic finance questions with weighted criteria, including a 300-question hard subset. An o4-mini judge scores the responses, so domain correctness and operational reliability remain separate questions.

1.Claude Fable 553.86%

2.Claude Opus 4.653.28%

3.GPT-551.32%

4.GPT-5 Pro51.06%

5.GPT-5.6 Sol50.45%

6.o3-pro49.08%

7.o347.69%

8.Claude Opus 4.546.16%

9.GPT-5.445.63%

10.Claude Sonnet 4.543.79%

11.Gemini 3.1 Pro Preview41.87%

12.o4-mini39.22%

13.Gemini 2.5 Pro38.92%

14.Gemini 2.5 Flash38.41%

15.GPT-4.134.32%

16.Llama 4 Maverick22.36%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-prbench-finance-full
Metric: weighted rubric score
Scale: 0-100
Participants: 16
Methodology and context: 600 finance tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 22.36% ± 0.42. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-70cdca115a0d. Source payload hash: 70cdca115a0da049a285a4ae00fcd400f49e1eea46970bf26fcd4c3374930b5c. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

PRBench Legal

Overall

500 realistic legal questions with weighted criteria across jurisdictions and practice areas. The benchmark measures professional response quality but cannot replace legal review or human accountability.

1.Claude Fable 552.56%

2.Claude Opus 4.652.27%

3.GPT-5.6 Sol50.5%

4.GPT-5 Pro49.89%

5.o3-pro49.67%

6.GPT-548.96%

7.o348.57%

8.GPT-5.444.35%

9.Claude Opus 4.544.21%

10.Gemini 3.1 Pro Preview44.02%

11.Gemini 2.5 Pro41.43%

12.Gemini 2.5 Flash41.02%

13.Claude Sonnet 4.540.84%

14.o4-mini38.11%

15.GPT-4.136.48%

16.Llama 4 Maverick24.84%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-prbench-legal-full
Metric: weighted rubric score
Scale: 0-100
Participants: 16
Methodology and context: 500 legal tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 24.84% ± 0.9. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-f4e7b1dd0a3e. Source payload hash: f4e7b1dd0a3ecd0a1d7d7b08e33377b62d408b58c2f0c61a413a4cfec0d57c2a. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

TutorBench

Overall

1,490 tasks test adaptive explanations, feedback, and hints across six STEM subjects. A Claude 4 Sonnet judge scores weighted rubrics, so strong results do not prove safe tutoring of real learners.

1.GPT-5.4 Pro56.62%

2.Gemini 2.5 Pro55.65%

3.GPT-555.33%

4.o3-pro54.62%

5.Gemini 3.1 Pro Preview52.99%

6.Gemini 3.1 Flash-Lite51.5%

7.Llama 4 Maverick40.2%

8.GPT-4o36.12%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-tutorbench-overall
Metric: weighted rubric score
Scale: 0-100
Participants: 8
Methodology and context: 1,490 tasks with weighted rubrics and Claude 4 Sonnet as judge.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 40.2% ± 1. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-642ef8e45495. Source payload hash: 642ef8e45495e730a3e2f2ddc114f60328e5c0c6da76f877e3b1435841792058. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

Vision

VISTA

Overall

758 image-text tasks combine OCR, spatial understanding, object recognition, and open-ended visual reasoning. The dataset is private and uses task-specific rubrics, making independent reproduction difficult.

1.Gemini 2.5 Pro54.63%

2.GPT-5.4 Pro53.89%

3.GPT-5 Pro52.39%

4.o3-pro51.63%

5.GPT-5.450.89%

6.GPT-5 mini50.39%

7.GPT-549.69%

8.GPT-4.145.34%

9.GPT-4.1 mini41.14%

10.Llama 4 Maverick38.33%

11.GPT-4.1 nano26.55%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-vista-private-overall
Metric: rubric score
Scale: 0-100
Participants: 11
Methodology and context: 758 single-turn image-text tasks with structured yes-no rubrics.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 38.33% ± 0.55. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-ca07b51d9088. Source payload hash: ca07b51d90881a89b710581569dec80da9f3d102bf237f2da6dbe4ad5c91b720. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

VisualToolBench

Overall

1,204 visual tasks require image manipulation, Python, web search, or calculator use with at most 20 tool calls. The score belongs to the complete multimodal tool system and cannot isolate the model alone.

1.GPT-5.429.17%

2.Gemini 3.1 Pro Preview28.97%

3.Claude Opus 4.627.52%

4.o313.74%

5.Gemini 2.5 Pro11.75%

6.o4-mini11.12%

7.GPT-4.15.52%

8.Gemini 2.5 Flash4.69%

9.Llama 4 Scout1.58%

10.Llama 4 Maverick1.41%

higher is betterAs of August 26, 2026Scale: 0% - 100%

Source: Scale AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: scale-visual-toolbench-apr
Metric: average pass rate
Scale: 0-100
Participants: 10
Methodology and context: 1,204 tasks, six tools, and at most 20 tool calls per task.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 1.41% ± 0.25. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-b6f662a26398. Source payload hash: b6f662a2639828660d3c7ee001d8c431eb80967131af7ab01b3e852f4e8f899d. Parser version: curated-v1. Imported at: 2026-09-29

Source: Scale AI

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

MMMU (Show measurement, test conditions, and source)

Source value: 73.4
Score: 73.4
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Meta Llama 4 model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Instruction-tuned table; BF16

Source Meta Llama 4 model card

MMMU Pro (Show measurement, test conditions, and source)

Source value: 59.6
Score: 59.6
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Meta Llama 4 model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Instruction-tuned table; BF16

Source Meta Llama 4 model card

MathVista (Show measurement, test conditions, and source)

Source value: 73.7
Score: 73.7
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Meta Llama 4 model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Instruction-tuned table; BF16

Source Meta Llama 4 model card

ChartQA (Show measurement, test conditions, and source)

Source value: 90
Score: 90
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Meta Llama 4 model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Instruction-tuned table; BF16

Source Meta Llama 4 model card

DocVQA (Show measurement, test conditions, and source)

Source value: 94.4
Score: 94.4
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Meta Llama 4 model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Instruction-tuned table; BF16

Source Meta Llama 4 model card

LiveCodeBench (Show measurement, test conditions, and source)

Source value: 43.4
Score: 43.4
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Meta Llama 4 model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Instruction-tuned table; BF16

Source Meta Llama 4 model card

Recorded measurement details

MultiNRC: Overall (Show measurement, test conditions, and source)

Source value: 8.44
Score: 8.44
Metric: accuracy
Unit: %
Benchmark version: French, Spanish, and Chinese release
Category: reasoning
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-multinrc-native-languages
Confidence interval: 1.68
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: Native, non-translated reasoning tasks with automatic evaluation.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 8.44% ± 1.68. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-bc7fb4bc0d9a. Source payload hash: bc7fb4bc0d9ad4757d46043b7c57b0b056967f2b0062249a99a3f0b616b95e23. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: Native, non-translated reasoning tasks with automatic evaluation.

Source Scale AI

PRBench Finance: Overall (Show measurement, test conditions, and source)

Source value: 22.36
Score: 22.36
Metric: weighted rubric score
Unit: %
Benchmark version: full finance set
Category: finance
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-prbench-finance-full
Confidence interval: 0.42
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 600 finance tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 22.36% ± 0.42. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-70cdca115a0d. Source payload hash: 70cdca115a0da049a285a4ae00fcd400f49e1eea46970bf26fcd4c3374930b5c. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 600 finance tasks with task-specific rubrics and o4-mini as judge.

Source Scale AI

PRBench Legal: Overall (Show measurement, test conditions, and source)

Source value: 24.84
Score: 24.84
Metric: weighted rubric score
Unit: %
Benchmark version: full legal set
Category: legal
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-prbench-legal-full
Confidence interval: 0.9
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 500 legal tasks with task-specific rubrics and o4-mini as judge.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 24.84% ± 0.9. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-f4e7b1dd0a3e. Source payload hash: f4e7b1dd0a3ecd0a1d7d7b08e33377b62d408b58c2f0c61a413a4cfec0d57c2a. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 500 legal tasks with task-specific rubrics and o4-mini as judge.

Source Scale AI

TutorBench: Overall (Show measurement, test conditions, and source)

Source value: 40.2
Score: 40.2
Metric: weighted rubric score
Unit: %
Benchmark version: 1,490-task release
Category: education
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-tutorbench-overall
Confidence interval: 1
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 1,490 tasks with weighted rubrics and Claude 4 Sonnet as judge.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 40.2% ± 1. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-642ef8e45495. Source payload hash: 642ef8e45495e730a3e2f2ddc114f60328e5c0c6da76f877e3b1435841792058. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 1,490 tasks with weighted rubrics and Claude 4 Sonnet as judge.

Source Scale AI

VISTA: Overall (Show measurement, test conditions, and source)

Source value: 38.33
Score: 38.33
Metric: rubric score
Unit: %
Benchmark version: 758 private prompt-image pairs
Category: multimodal
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-vista-private-overall
Confidence interval: 0.55
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 758 single-turn image-text tasks with structured yes-no rubrics.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 38.33% ± 0.55. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-ca07b51d9088. Source payload hash: ca07b51d90881a89b710581569dec80da9f3d102bf237f2da6dbe4ad5c91b720. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 758 single-turn image-text tasks with structured yes-no rubrics.

Source Scale AI

VisualToolBench: Overall (Show measurement, test conditions, and source)

Source value: 1.41
Score: 1.41
Metric: average pass rate
Unit: %
Benchmark version: 1,204-task release
Category: multimodal
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: scale-visual-toolbench-apr
Confidence interval: 0.25
Source type: official-leaderboard
Evaluator: Scale AI
Status: active
Retrieved at: 2026-08-26
Methodology: 1,204 tasks, six tools, and at most 20 tool calls per task.. Source model ID: curated/llama-4-maverick. Source model name: Llama 4 Maverick. Provider reported by source: Meta. Evaluator: Scale AI. Dataset type: public. Attempts: pass@1. Score display: 1.41% ± 0.25. Source ID: scale-ai. Evidence capture ID: curated-scale-ai-2026-08-26-b6f662a26398. Source payload hash: b6f662a2639828660d3c7ee001d8c431eb80967131af7ab01b3e852f4e8f899d. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 1,204 tasks, six tools, and at most 20 tool calls per task.

Source Scale AI

Page 1 of 3

Predecessor and successor

Sources and data date

Type	Evidence and data date
Speed	Median across serving providers. Output speed after the first streaming chunk, not total response time. Retrieved September 8, 2026. Source: Artificial Analysis · Artificial Analysis (retrieved September 8, 2026)Speed measurement methodology
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · API pricing
Input	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Meta Llama 4 model card (retrieved September 8, 2026) · Meta Llama 4 model card · Editorial description reviewed October 2, 2026
Output	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Meta Llama 4 model card (retrieved September 8, 2026) · Meta Llama 4 model card · Editorial description reviewed October 2, 2026
Research status	Research date September 8, 2026. 2 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources	max-output-tokens Original wording Card does not specify a model-specific maximum output.
Additional source	Model metadata
Additional source	Context window data
Additional source	Knowledge cutoff data · Meta AI, The Llama 4 herd · Editorial description reviewed October 2, 2026
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	Scale AI (retrieved August 26, 2026)
Additional source	OpenRouter: Popularity index (derived), CC BY 4.0
Additional source	Vercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional source	Creative Commons Attribution 4.0

Llama 4 Maverick

Specifications and access

Published benchmarks

MultiNRC

PRBench Finance

PRBench Legal

TutorBench

Vision

VISTA

VisualToolBench

Measurements without matching peer values

Recorded measurement details

Our tests

Pelican on a bicycle

Popularity

Published prices

Predecessor and successor

Sources and data date