Hands-on with Google's MKA multimodal benchmark across six AI workflows
Emergency_Abies_6508 · reddit · 2026-11-13
The author ran Google and Artificial Analysis' new Multimodal Knowledge Agent (MKA) benchmark themselves: they downloaded an Education/History task from Hugging Face and tested it across six AI workflows including GPT, Gemini, Claude, and NotebookLM, grading outputs against the official rubric.
Findings:
- Some tools produced great-looking presentations but stumbled on getting the numbers right (e.g., castle and settlement statistics for Guadalajara and Málaga shown in the sample slides)
- A blog post documents the results, actual answers, and observations
A practical first-hand comparison for anyone curious how multimodal agents differ on real knowledge tasks.
More from Models
- Chinese major models reportedly trained on Blackwell-class chips, though scale disappoints — teortaxesTex · 2026-11-13
- One user burned $4,988 of API-priced tokens on a $200 Claude subscription last month — jonathan_wilke · 2026-11-13
- Gemini 4 Hydrogen tops Google models on AA-Robotics v0.1, ranking fourth overall — ArtificialAnlys · 2026-11-13
- 4,689-trial agent test: DeepSeek V4 argues its way into failures, Gemini 3.8 Flash fails silently — Kindly-Bag763 · 2026-11-13
- Analyst: DeepSeek V4's Flash variant is winning, and a hyper-efficient 2B V5 could be the real play — teortaxesTex · 2026-11-13
- Interaction-as-a-Judge beats code rewards: +6.7pp win rate, 35.7% less code bloat — arena · 2026-11-13