RLTasksby ulam.ai
Models / Google

Google / MODEL EVIDENCE

Gemini-3 Pro

29 benchmarks · 34 observations · 5 sources · released 2025-11-18

Scores at a glance

29 benchmarks · large dot: Gemini-3 Pro; small dots: closest models where reported

MCPMark ↗53.9
29.557.5

Gemini-3 Pro · MCP-Mark · original release · Qwen report ↗

MMMU ↗87.2
80.687.2

Gemini-3 Pro · MMMU · Qwen report ↗

OmniDocBench ↗92.91 points
78.4496.91

Gemini 3 Pro · OmniDocBench · Overall parsing score · end-to-end evaluation ↗

UnsolvedMath ↗5 / 5
55

Gemini 3 Pro · UnsolvedMath · Literature search · Ulam · January 31, 2026 research assessment ↗

3 more reported results

2 / 5 · Gemini 3 Pro · UnsolvedMath · Theorem proving · Ulam · January 31, 2026 research assessment · Open Mathematical Problems as an AI Reasoning Benchmark ↗

1 / 5 · Gemini 3 Pro · UnsolvedMath · Proof checking · Ulam · January 31, 2026 research assessment · Open Mathematical Problems as an AI Reasoning Benchmark ↗

0 / 1 · Gemini 3 Pro · UnsolvedMath · Formalization · Ulam · January 31, 2026 research assessment · Open Mathematical Problems as an AI Reasoning Benchmark ↗

Closest models by score

Similar positions on shared benchmark leaderboards. Smaller gaps are closer.

42.8 percentile-point gap

26 of 28 shared benchmark comparisons

45.0 percentile-point gap

26 of 28 shared benchmark comparisons

How similarity is measured

Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.

Filter comparisons to this model →

Reports featuring this model

Qwen3.5 · earlier-2026 baseline · Qwen →

Sources

Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.

  1. Vending Bench 2 — methodology and published results ↗

    Epoch AI · 2026

  2. AI benchmarks & capabilities ↗

    Epoch AI

  3. Qwen/Qwen3.5-397B-A17B · Hugging Face ↗

    Qwen Team

  4. OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations — published result table ↗

    Linke Ouyang, Yuan Qu, Hongbin Zhou, Jiawei Zhu, Rui Zhang, Qunshu Lin et al. · 2024

    All 20 authors

    Linke Ouyang, Yuan Qu, Hongbin Zhou, Jiawei Zhu, Rui Zhang, Qunshu Lin, Bin Wang, Zhiyuan Zhao, Man Jiang, Xiaomeng Zhao, Jin Shi, Fan Wu, Pei Chu, Minghao Liu, Zhenxiang Li, Chao Xu, Bo Zhang, Botian Shi, Zhongying Tu, Conghui He

  5. Open Mathematical Problems as an AI Reasoning Benchmark ↗

    Przemyslaw Chojecki · 2026

Names used in source tables

Gemini 3 Pro · Gemini-3 Pro