Gemini 3 Pro Preview

What the data says

Benchmarked on 14 of 27 tracked axes. Every figure below is the model's own published result on that board.

Uneven: 82.9 percentile points between its best and worst axis — 100 pct on MMLU-Pro (Artificial Analysis) versus 17.1 pct on HHEM Hallucination (Vectara).

• MMLU-Pro (Artificial Analysis)100 pct89.8%Verify →
• LiveCodeBench (Artificial Analysis)100 pct91.7%Verify →
• BFCL v4 (function calling)Dormant board97.6 pct72.5%Verify →
• SWE-bench Verified (live board)95.4 pct77.4%Verify →
• SimpleBench94.3 pct76.4%Verify →
• GPQA Diamond90 pct92.6%Verify →
• Terminal-Bench86.4 pct69.4%Verify →
• Humanity's Last Exam84.8 pct37.5%Verify →
• FrontierMath84.1 pct37.6%Verify →
• IFBench (Artificial Analysis)82.7 pct70.4%Verify →
• AIME (math)74 pct91.4%Verify →
• ARC-AGI-253.9 pct31.1%Verify →
• SWE-bench Verified37.5 pct72.9%Verify →
• HHEM Hallucination (Vectara)17.1 pct86.4%Verify →
• OverallNot benchmarked
• WebDev ArenaNot benchmarked
• Vision ArenaNot benchmarked
• Factuality ArenaNot benchmarked
• Search ArenaNot benchmarked
• Document ArenaNot benchmarked
• Agent ArenaNot benchmarked
• SimpleQA VerifiedNot benchmarked
• LiveBenchNot benchmarked
• Terminal-Bench 4.0Not benchmarked
• CodingNot benchmarked
• Instruction followingNot benchmarked
• Hard promptsNot benchmarked

Percentiles are the model's standing within each board's own population. Axes marked "not benchmarked" have no published result for this model — that is an absence of data, not a low score.