Gemini Pro · 发布历史
Google Gemini Pro 系列已收录 3 次发布和 61 个测试版本。
Gemini Pro 主要型号版本;同版有依据的配置分别保留,其他产品家族不混入。
全部版本
- Gemini 3.1 Pro18 项测试
- Gemini 3 Pro26 项测试
- Gemini 2.5 Pro (GA)35 项测试
已收录测试 · 61
- ARC-AGI-2 · ARC Prize Verified · semi-private setGemini 3.1 Pro、Gemini 3 Pro
- Humanity's Last Exam · 2,500-question benchmarkGemini 3.1 Pro、Gemini 3 Pro、Gemini 2.5 Pro (GA)
- GPQA Diamond · 198-question Diamond subsetGemini 3.1 Pro、Gemini 3 Pro、Gemini 2.5 Pro (GA)
- Terminal-Bench · 2.0;代理框架按协议区分Gemini 3.1 Pro、Gemini 3 Pro
- SWE-bench Verified · Verified subset (500 problems)Gemini 3.1 Pro、Gemini 3 Pro、Gemini 2.5 Pro (GA)
- SWE-Bench Pro · PublicGemini 3.1 Pro
- LiveCodeBench Pro · Competitive coding problems from Codeforces, ICPC, and IOIGemini 3.1 Pro、Gemini 3 Pro
- SciCode · Scientific research codingGemini 3.1 Pro
- APEX-Agents · Long-horizon professional tasksGemini 3.1 Pro
- GDPval-AA · Artificial Analysis leaderboardGemini 3.1 Pro
- τ²-bench · Retail sectionGemini 3.1 Pro、Gemini 3 Pro
- τ²-bench · Telecom sectionGemini 3.1 Pro、Gemini 3 Pro
- MCP Atlas · Public setGemini 3.1 Pro
- BrowseComp · 1,266 题;单代理与多代理按协议区分Gemini 3.1 Pro
- MMMU-Pro · Standard and Vision formatsGemini 3.1 Pro、Gemini 3 Pro
- MMMLU · Multilingual Q&AGemini 3.1 Pro、Gemini 3 Pro
- MRCR-V2 · 8-needle; up to 128K average contextGemini 3.1 Pro、Gemini 3 Pro、Gemini 2.5 Pro (GA)
- MRCR-V2 · 8-needle; pointwise at 1M contextGemini 3.1 Pro、Gemini 3 Pro、Gemini 2.5 Pro (GA)
- LMArena · Gemini 3 release leaderboard scoreGemini 3 Pro
- MathArena Apex · MathArena leaderboardGemini 3 Pro
- Video-MMMU · Gemini 3 Pro model evaluation; recommended media_resolution=HIGH and temperature=0Gemini 3 Pro
- SimpleQA Verified · Kaggle official leaderboardGemini 3 Pro
- WebDev Arena · Gemini 3 release leaderboard scoreGemini 3 Pro
- Vending-Bench · 2Gemini 3 Pro
- AIME · 2025 editionGemini 3 Pro、Gemini 2.5 Pro (GA)
- CharXiv Reasoning · Validation splitGemini 3 Pro
- OmniDocBench 1.5 · Version 1.5; average edit distance across text, formula, table and reading orderGemini 3 Pro
- tau2-bench · Average across Retail, Airline and Telecom categoriesGemini 3 Pro
- FACTS Benchmark Suite · Held-out internal grounding, parametric, multimodal and search-retrieval benchmarksGemini 3 Pro
- Global PIQA · Commonsense reasoning across 100 languages and culturesGemini 3 Pro
- tau2-bench · Airline categoryGemini 3 Pro
- ScreenSpot-Pro · 1,581 expert-annotated tasks across 23 professional applications and 3 operating systemsGemini 3 Pro
- LiveCodeBench · UI 2025-01-01 to 2025-05-01Gemini 2.5 Pro (GA)
- Aider Polyglot · Diff or diff-fenced edit formatGemini 2.5 Pro (GA)
- SimpleQA · No search-enabled; report F1Gemini 2.5 Pro (GA)
- FACTS Grounding · 2024 releaseGemini 2.5 Pro (GA)
- Global MMLU Lite · Lite; 15 human-translated languagesGemini 2.5 Pro (GA)
- ECLeKTic · Closed-book QAGemini 2.5 Pro (GA)
- HiddenMath-Hard · Hard held-out setGemini 2.5 Pro (GA)
- MMMU · Single-attemptGemini 2.5 Pro (GA)
- Vibe-Eval · Reka image-understanding setGemini 2.5 Pro (GA)
- ZeroBench · Average over 4 runsGemini 2.5 Pro (GA)
- BetterChartQA · 9 capability bucketsGemini 2.5 Pro (GA)
- FLEURS · 53-language subsetGemini 2.5 Pro (GA)
- CoVoST2 · 21 languages to EnglishGemini 2.5 Pro (GA)
- ActivityNet-QA · Test subsetGemini 2.5 Pro (GA)
- EgoTempo · Test subsetGemini 2.5 Pro (GA)
- Perception Test · Test subsetGemini 2.5 Pro (GA)
- QVHighlights · Validation subset; [email protected]Gemini 2.5 Pro (GA)
- Video-MMMU · Test subsetGemini 2.5 Pro (GA)
- 1H-VideoQA · Test subsetGemini 2.5 Pro (GA)
- LVBench · Test subsetGemini 2.5 Pro (GA)
- VideoMME · Audio + visual; Long test subset; 0-shot; max 1024 frames at 1 fpsGemini 2.5 Pro (GA)
- VATEX · Test subsetGemini 2.5 Pro (GA)
- VATEX-ZH · Validation subsetGemini 2.5 Pro (GA)
- YouCook2Cap · Validation subsetGemini 2.5 Pro (GA)
- Minerva · Video; test subsetGemini 2.5 Pro (GA)
- Neptune · Video; test subsetGemini 2.5 Pro (GA)
- VideoMME · Full test set; audio + visual + subtitlesGemini 2.5 Pro (GA)
- LOFT · Hard retrieval; average context length up to 128K; 300 queriesGemini 2.5 Pro (GA)
- LOFT · Hard retrieval; 1M context, pointwise valueGemini 2.5 Pro (GA)