跳轉到主要內容

Muse Spark

MetaProprietary

描述

Muse Spark is the first model in the Muse family developed by Meta Superintelligence Labs. It is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration. It features a Contemplating mode that orchestrates multiple agents reasoning in parallel. It demonstrates competitive performance in multimodal perception, reasoning, health, and agentic tasks, with Contemplating mode achieving 58% on Humanity's Last Exam and 38% on FrontierScience Research.

發布日期
2026-04-08
參數規模
上下文長度
支援模態

能力雷達圖

42
general
58
coding
88
reasoning
66
science估算
80
agents
70
multimodal

Science 在缺少專門科學評測時使用推理能力代理估算。

排行榜排名

領域#排名分數來源
智慧體能力模型榜67
54.0
LS
程式碼能力榜33
79.0
AA
通用能力榜19
81.0
AA
多模態榜77
60.0
LS
推理能力92
50.0
LS
科學能力15
84.0
AA

基準測試分數 (LLM Stats)

Agents

GDPval-AA1164.00 / 3000自報
DeepSearchQA74.8%自報
Terminal-Bench 2.059.0%自報
SWE-Bench Pro52.4%自報

Biology

GPQA89.5%自報

Code

LiveCodeBench Pro0.80 / 3000自報
SWE-Bench Verified77.4%自報

Communication

Tau2 Telecom91.5%自報

General

MMMU-Pro80.4%自報
SimpleVQA0.71 / 100自報

Grounding

ScreenSpot Pro84.1%自報

Healthcare

MedXpertQA78.4%自報
HealthBench Hard42.8%自報

Math

Humanity's Last Exam58.4%自報

Multimodal

CharXiv-R86.4%自報
ZEROBench0.33 / 100自報

Physics

IPhO 202582.6%自報

Reasoning

ERQA64.7%自報
ARC-AGI v242.5%自報
FrontierScience Research38.3%自報

AA 評測指數

Coding Index
58.6
Intelligence Index
43.1
Tau2
0.9
Gpqa
0.9
Ifbench
0.8
Lcr
0.7
Terminalbench V2 1
0.6
Scicode
0.5
Terminalbench Hard
0.5
Hle
0.4
Tau Banking
0.2

LLM Stats 分類評分

Legal
100
Finance
100
Agents
100
General
100
Reasoning
78
Physics
90
Biology
90
Chemistry
90
Communication
90
Frontend Development
80
Grounding
80
Tool Calling
80
Image To Text
70
Multimodal
70
Search
70
Code
70
Vision
70
Math
60
Spatial Reasoning
60
Healthcare
60

定價

輸入價格免費
輸出價格免費
混合價格(3:1)免費

速度

Tokens/秒0.0
首Token延遲0.00s
首回答延遲0.00s

供應商價格排行

暫無提供商資料

外部連結