Best models for multimodal understanding

MKA

Multimodal Climb weighted-success-rate accuracy, independently measured at the recorded model effort.

Updated: Nov 12, 2026Data as of: Nov 12, 2026

Limited reportingNot saturated

MKA results. Each row identifies its benchmark version and evaluation track. Scores from different versions are not directly comparable. Best effort from the latest stored report date per model and version.
Model / creator	Version	Reported score
Gemini 4 HydrogenGoogle · Proprietary	Climb WSR / Google Hydrogen launch	53.4%
Claude Sonnet 5.5Anthropic · Proprietary	Climb WSR / Google Hydrogen launch	52%
Gemini 4 HydrogenGoogle · Proprietary	Climb WSR / independent xhigh	49.52%
GPT-6.1 TerraOpenAI · Proprietary	Climb WSR / Google Hydrogen launch	43.8%

View all 4 published reports

Gemini 4 Hydrogen49.52%

Climb WSR / independent xhigh

effort: xhigh · Exact independent xhigh record; methodology per published evaluation.

Gemini 4 Hydrogen53.4%

Climb WSR / Google Hydrogen launch

effort: highest available or best reported · Hydrogen self-computed by Google; peer figures quoted from independent evaluation. Not the separate xhigh independent Hydrogen record. · No tools

Google DeepMind ↗

Claude Sonnet 5.552%

Climb WSR / Google Hydrogen launch

effort: highest available or best reported · Hydrogen self-computed by Google; peer figures quoted from independent evaluation. Not the separate xhigh independent Hydrogen record. · No tools

Google DeepMind ↗

GPT-6.1 Terra43.8%

Climb WSR / Google Hydrogen launch

effort: highest available or best reported · Hydrogen self-computed by Google; peer figures quoted from independent evaluation. Not the separate xhigh independent Hydrogen record. · No tools

Google DeepMind ↗

Sources and original reports ↗

Other benchmarks

LiveBench: Coding

Tests coding tasks, helping you compare models for implementation work that you would otherwise write or debug yourself.

Agents’ Last Exam

Evaluates completion of long professional workflows with graded deliverables.

FrontierCode

Tests whether an agent can produce software changes that meet functional and review requirements, helping you assess how much implementation work you could delegate before human code review.