Google DeepMind releases MKA benchmark for multimodal agent tasks
With @ArtificialAnlys, we built MKA (Multimodal Knowledge Agent): a benchmark for how well agents ha...
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.
Google DeepMind, together with Artificial Analysis, has introduced MKA (Multimodal Knowledge Agent), a benchmark that measures how well AI agents handle visually complex tasks. On the first published ranking, Gemini 4 Carbon places #2 while running at 6x lower average cost per task than the top-ranked model. The benchmark is available via a link shared in DeepMind's announcement.
MKA targets a gap the team says existing evaluations do not cover: professional work that goes beyond text, such as cross-referencing blueprints or reading scientific charts. Rather than testing single-image understanding, it evaluates agentic performance on tasks where visual information must be combined, verified and acted on across documents.
The result positions Gemini 4 Carbon as a cost-efficiency leader in agentic multimodal work, even without holding the top accuracy spot. For enterprises evaluating agents for document-heavy workflows like engineering review or research analysis, cost per task at comparable capability levels is becoming a key differentiator alongside raw benchmark scores.
With @ArtificialAnlys, we built MKA (Multimodal Knowledge Agent): a benchmark for how well agents ha...
With @ArtificialAnlys , we built MKA (Multimodal Knowledge Agent): a benchmark for how well agents handle visually complex tasks. Everyday professional work often uses more than just text – from cross-referencing blueprints to reading scientific charts. This is not covered by existing evaluations. Gemini 4 Carbon ranks #2 , at 6x lower average cost per task than the top model. → goo.gle/4AikIja
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Gemini 4 Carbon, Google DeepMind, Artificial Analysis, MKA benchmark.
Source attribution: curated from Google DeepMind (@GoogleDeepMind).
