What happened
Artificial Analysis and Google DeepMind launched the Multimodal Knowledge Agent (MKA) benchmark, with Anthropic's Claude Fable 5.2 (max) leading at a Climb WSR of 57%, followed by Gemini 4 Carbon (high) at 54% and Claude Opus 5.5 (max) at 53%.
Why it matters
The benchmark scores agents on producing deliverables like slide decks and spreadsheets from dense visual files, where no single model leads across all 10 industry verticals.
WHO IT HITSEnterprise teams evaluating AI agents for document-heavy knowledge work, such as engineering or healthcare, may get a clearer view of which models handle real-world visual inputs.
Summaries like this, every morning.
The Multimodal Knowledge Agent (MKA) benchmark arrives as AI agents move from chat toward producing finished work products. Its 200 private real-world visual tasks across 10 industries are built from dense project files such as architectural drawings, engineering schematics, and clinical scans, which the benchmark says are genuinely required to complete the task rather than just added context.
Scoring separates Regression checks, baseline criteria models should pass, from Climb criteria designed to separate stronger models. The headline Climb WSR number therefore reflects the harder end of the rubric. Evaluation runs in Stirrup, an open-source agent harness with code execution, web search, web fetch, and image-viewing tools up to 250 turns, and deliverables are graded by GPT-6 Luna (medium) as an agentic judge.
Performance is uneven across domains and output types. Claude Fable 5.2 leads five industry verticals, while Claude Opus 5.5 (max) takes Education and Construction, Gemini 4 Argon (high) takes Science, Gemini 4 Carbon (high) takes Healthcare, and Claude Sonnet 5.5 (max) narrowly leads Law & Government. Word Document and Text File outputs are the hardest, with no model passing 52% on either.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
