# Google and Artificial Analysis launch the Multimodal Knowledge Agent benchmark

> Original title: Introducing the Multimodal Knowledge Agent benchmark

- Source: Google · Gemini app
- Published: 2026-11-10T16:00:00.000Z
- Bad Influence: https://badinfluence.news/items/svco2mfo6ez9dtfive76fby2c
- Original: https://blog.google/innovation-and-ai/models-and-research/gemini-models/multimodal-knowledge-agent-benchmark/

## Summary

Google and Artificial Analysis launch MKA, a benchmark that evaluates how agents reason across visually complex files on real-world knowledge work tasks. The 200-task evaluation set is kept private, and a separate 10-task public set is released on Hugging Face. Google says Gemini 4 Carbon ranks #2 in quality among the tested models at 6x lower average cost than the top scorer.