跳轉到主要內容

Llama 4 Maverick

MetaLlama開源權重Llama 4 Community License Agreement

描述

Llama 4 Maverick is a natively multimodal model capable of processing both text and images. It features a 17 billion active parameter mixture-of-experts (MoE) architecture with 128 experts, supporting a wide range of multimodal tasks such as conversational interaction, image analysis, and code generation. The model includes a 1 million token context window.

發布日期
2025-04-05
參數規模
400.0B
上下文長度
1.0M
支援模態
image, text

能力雷達圖

31
general
26
coding
39
reasoning
42
science
35
agents
90
multimodal

排行榜排名

領域#排名分數來源
程式碼能力榜431
30.0
AA
通用能力榜385
35.0
AA
多模態榜104
45.0
LS
科學能力391
38.0
AA

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

General

MMLU85.5%自報

Language

MMLU-Pro80.5%自報
TydiQA31.7%自報

Math

MGSM92.3%自報
MathVista73.7%自報
MATH61.2%自報

Multimodal

MMMU73.4%自報

Reasoning

ChartQAMasry et al. (2022)90.0%自報
MBPP0.78 / 100自報
GPQANYU + Cohere + Anthropic (2023)69.8%自報
LiveCodeBench43.4%自報

Vision

DocVQADocVQA (2020)94.4%自報
MMMU-Pro59.6%自報

AA 評測指數

(Artificial Analysis)
Math 500(OpenAI (2024), subset of Hendrycks et al. MATH (2021))
88.9
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))
80.9
Gpqa(NYU + Cohere + Anthropic (2023))
67.1
Lcr(Artificial Analysis)
50.0
Ifbench(Google Research (2023))
43.0
Livecodebench(UC Berkeley + MIT + Cornell (2024))
39.7
Aime(MAA (Mathematical Association of America))
39.0
Scicode(UIUC + Argonne National Lab (2024))
31.7
Aime 25(MAA (Mathematical Association of America))
19.3
Math Index(Artificial Analysis)
19.3
Tau2(Sierra + U Toronto + Vector Institute (2025))
17.8
Coding Index(Artificial Analysis)
16.3
Intelligence Index(Artificial Analysis)
10.0
Terminalbench V2 1
7.9
Terminalbench Hard(Stanford × Laude Institute (2026))
6.8
Hle(Center for AI Safety + Scale AI (2025))
4.9
Tau Banking
3.7
Terminalbench V4 0
0.0

LLM Stats 分類評分

(LLM Stats (zeroeval))
Image To Text
90
Language
80
Legal
80
Math
80
Multimodal
80
Finance
80
General
80
Healthcare
80
Vision
80
Physics
70
Reasoning
70
Biology
70
Chemistry
70
Code
40

定價

輸入價格$0.26 / 1M tokens
輸出價格$0.91 / 1M tokens
混合價格(3:1)$0.422 / 1M tokens

速度

Tokens/秒94.1
首Token延遲0.58s
首回答延遲0.58s

供應商價格排行

供應商價格排行

8 個供應商

最便宜: DeepInfra最貴: Neon
供應商輸入輸出
1DeepInfra最便宜
$0
$0
2NanoGPT
$0.15
$0.6
3Helicone
$0.15
$0.6
4OpenRouter
$0.1875
$0.6525
5Kilo Gateway
$0.1875
$0.6525
6DigitalOcean
$0.25
$0.87
7Meta主要
$0.26
$0.91
8Neon
$0.5
$1.5

比較該模型在不同 API 供應商之間的定價。

外部連結