evals.report
BenchmarksLabsCompareRun guidesIn the wild

Gemini 3 Deep Think

Google DeepMind · Gemini. Released Dec 4, 2025.

Gemini 3 Deep Think is a model from Google DeepMind in the Gemini family, released Dec 4, 2025. evals.report tracks 4 reported Gemini 3 Deep Think benchmark scores across ARC-AGI-1, ARC-AGI-2, LiveCodeBench Pro, GDPval — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

4 results

Benchmark results 4

Compare this model
BenchmarkCategoryScoreMetricStatusDate
ARC-AGI-1Reasoning96%accuracyOfficialDec 4, 2025Details
ARC-AGI-2Reasoning84.58%accuracyOfficialDec 4, 2025Details
LiveCodeBench ProCoding3298Codeforces EloOfficialDec 4, 2025Details
GDPvalAgents1324EloOfficialDec 4, 2025Details