跳轉到主要內容

Muse Glimmer-30B

Meta開源權重Apache 2.0 · 商用許可

描述

Muse Glimmer-30B is Meta Superintelligence Labs' open-weight, dense multimodal model for autonomous agentic work on consumer hardware. Its 29.6-billion-parameter causal transformer includes a roughly 1.8B-parameter ViT-G/14 perception encoder, accepts interleaved text and images, produces text, and supports a 131,072-token context window. It combines controllable multi-step reasoning, schema-based tool use, coding, failure recovery, and compatibility with agent scaffolds such as OpenClaw and Hermes Agent. Meta also releases 4-bit variants sized for 24 GB and 32 GB systems and a DFlash speculative-decoding drafter. The model is trained on more than 100 languages and released under Apache 2.0.

發布日期
2026-08-10
參數規模
29.6B
上下文長度
131K
支援模態
image, text

能力雷達圖

60
general
60
coding
50
reasoning
51
science估算
50
agents
70
multimodal

缺少專門科學評測時,Science 由 LLM Stats 科學得分或推理能力估算。

排行榜排名

領域#排名分數來源
智慧體能力模型榜132
33.0
LS
多模態榜52
45.0
LS

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Agents

Siren AgentDojo Utility94.2%自報
MCP Atlas75.5%自報
DeepSearchQA74.6%自報
OSWorld-Verified65.9%自報
CI Memories Coverage64.8%自報
Terminal-Bench 2.151.7%自報
SWE-Bench ProPrinceton NLP (2024)51.2%自報
WildClawBench47.6%自報
SkillsBench44.3%自報
GAIA243.3%自報
Siren AgentDojo Attack Success Rate28.4%自報
CI Memories Violation Rate26.4%自報
Tau3 Banking23.5%自報

Biology

GPQANYU + Cohere + Anthropic (2023)83.5%自報
SciCode43.6%自報

Code

SWE-Bench Verified76.0%自報

General

IFBench77.0%自報
MMMU-Pro74.0%自報

Grounding

ScreenSpot Pro75.4%自報

Long Context

AA-LCR80.0%自報
Beam 128K65.1%自報

Math

AIME 202694.7%自報
Humanity's Last Exam22.0%自報

Multimodal

CharXiv-R78.8%自報
OmniDocBench 1.575.8%自報

AA 評測指數

(Artificial Analysis)

暫無 AA 評測資料

LLM Stats 分類評分

(LLM Stats (zeroeval))
Long Context
80
Spatial Reasoning
80
Structured Output
80
Instruction Following
80
Frontend Development
80
Grounding
80
Multimodal
70
Search
70
Vision
70
Physics
60
Reasoning
60
General
60
Biology
60
Chemistry
60
Code
60
Math
50
Agents
50
Coding
50
Tool Calling
50

定價

輸入價格$0.35 / 1M tokens
輸出價格$1.5 / 1M tokens
混合價格(3:1)$0.6375 / 1M tokens
快取讀取價格$0.04 / 1M tokens

速度

暫無速度資料

供應商價格排行

供應商價格排行

5 個供應商

最便宜: Kilo Gateway最貴: Vercel AI Gateway
供應商輸入輸出
1Kilo Gateway最便宜
$0.3
$1.2
2Meta主要
$0.35
$1.5
3NanoGPT
$0.35
$1.5
4OpenRouter
$0.35
$1.5
5Vercel AI Gateway
$0.35
$1.5

比較該模型在不同 API 供應商之間的定價。

外部連結