AITOPAI热报
Models

与 Artificial Analysis 联合发布多模态知识智能体基准 MKA

Patrick Loeber@patloeber
Post
View the post
Why we picked it

做文档智能体或 RAG 的朋友可以看看这个新基准，200 个任务横跨 10 个行业，专测模型读复杂文件再干活的能力，比单轮问答基准更贴近实际工作流，选模型前先去 Artificial Analysis 看排名。

双方联合推出名为 Multimodal Knowledge Agent（MKA）的新基准，用于测试智能体在复杂视觉文件上的知识型工作能力。

Body

MKA 包含 200 个任务，覆盖 10 个行业垂直领域，任务形式要求模型在读取和理解复杂视觉内容（如文档、图表类文件）的基础上完成多步骤的知识工作。评测结果已在 Artificial Analysis 的评估页面（artificialanalysis.ai/evaluations/mka）公开，可直接查看各模型的得分对比。

对开发者而言，MKA 把评测焦点从单轮多模态问答转向多步骤智能体工作流，可以作为选型时衡量模型在真实文件处理场景表现的参考；此前常见的基准多停留在 OCR 或 VQA 类单步任务，MKA 的垂直行业划分也更贴近企业实际使用场景。

Original · Patrick Loeber

we partnered with @ArtificialAnlys to launch a new benchmark: Multimodal Knowledge Agent (MKA) It tests agentic knowledge work over complex visual files, including 200 tasks across 10 industry verticals. Results: artificialanalysis.ai/evaluations/mka📊 22 ⚡

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include MKA, Artificial Analysis, 多模态, 智能体.

Source attribution: curated from Patrick Loeber (@patloeber).

#MKA#Artificial Analysis#多模态#智能体#基准测试

Keep reading

Recent stories with the same tag or section
ModelsUS vs China07:28
Chinese contentMKA 基准发布：针对视觉复杂文件的知识工作智能体评测

Artificial Analysis 和 Google 做了个专门测智能体处理复杂文档的基准，任务由行业专家写，还故意保密防止模型背题。首批数据里 Gemini 4 Carbon 分数略输 Claude Fable 5.2，但成本只有对方五分之一，做模型选型的人可以直接看榜单。

#MKA#Artificial Analysis#Gemini 4 Carbon#Claude
Philipp Schmid@_philschmidSource
ModelsBuildersSelected02:21
Google DeepMind releases MKA benchmark for multimodal agent tasks

A #2 ranking paired with 6x lower cost per task signals that the agent market is shifting from accuracy-only leaderboards to cost-adjusted comparisons, a framing Artificial Analysis is actively promoting. Teams procuring agents for document-heavy work in engineering, science or operations should track MKA scores when shortlisting models. Buyers in ASEAN and global markets can use the cost-per-task metric to compare against pricier frontier alternatives before committing to long-context multimodal deployments.

#Gemini 4 Carbon#Google DeepMind#Artificial Analysis#MKA benchmark
Official accountGoogle DeepMind@GoogleDeepMindSource
ModelsUS vs China01:06
Chinese contentACE Music v1 发布即登顶 Artificial Analysis 人声音乐榜，超过 Suno v6

中国团队 ACE Studio 的音乐模型刚发 v1 就在人声榜单上压过 Suno v6，这个对比本身就很值得看。支持自己喂歌词、做纯伴奏，雷鬼和古典管弦乐是强项。每天 20 积分免费用，做歌的人可以拿去和 Suno 同一首提示词对比下效果。

#ACE Music#Suno#ACE Studio#音乐生成
berryxia@berryxiaSource
ModelsUS vs China01:05
Chinese content百度开源 ERNIE 5.5 Flash Preview：280B 总参、14B 激活，办公自动化榜超 Claude Opus 5.5

百度这个 ERNIE 5.5 Flash 用 14B 激活参数在办公自动化和表格任务上压过了 Claude Opus 5.5，但代码分数还差不少，而且榜单是百度自己测的。等你需要跑表格和流程自动化，11 月 25 日前权重放出后可以部署对比下；纯写代码的话先留在 Claude。

#ERNIE 5.5 Flash#百度#Claude Opus 5.5#开源模型
berryxia@berryxiaSource
ModelsUS vs ChinaSelected01:00
ACE Music v1 tops Artificial Analysis vocal music leaderboard

The ranking suggests Chinese AI music tooling has caught up with Western leaders on vocal quality, a domain where Suno previously set the benchmark. Producers and indie musicians can try it immediately through the app's free tier, though the lack of an API limits integration into automated workflows. Teams comparing music generators should run their own genre tests, since performance varies by style.

#ACE Music#ACE Studio#Suno#music generation
Official accountArtificial Analysis@ArtificialAnlysSource
ModelsBuilders79°01:00
Chinese contentGemini Omni Flash 1.2 视频生成评测：AA-Video-Editing 登顶第二，价格降 31%

Google 这款新视频模型编辑能力排到榜单第二，1080p 每分钟 6.26 美元，比上代便宜三成。做短视频、UI 动效或视频特效的团队可以直接在 Gemini API 和 Google Flow 里试，12 月 3 日它就替换掉旧版 1.1 了。

#Gemini#Google#视频生成#多模态
Official accountArtificial Analysis@ArtificialAnlysSource