evals.report
BenchmarksLabsCompareRun guidesIn the wild
Google DeepMindGemini

Gemini 2.5 Flash

Google DeepMind · Gemini. Released Apr 17, 2025.

Gemini 2.5 Flash is a model from Google DeepMind in the Gemini family, released Apr 17, 2025. evals.report tracks 27 reported Gemini 2.5 Flash benchmark scores across Berkeley Function Calling Leaderboard, FrontierMath, AIME (OTIS Mock), LiveCodeBench Pro, Humanity's Last Exam, LMArena, WeirdML, MCP Atlas, and 19 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

27 results

Benchmark results 27

Compare this model
BenchmarkCategoryScoreMetricStatusDate
Berkeley Function Calling LeaderboardTool use56.24%accuracyOfficialApr 17, 2025Details
FrontierMathReasoning4.84%accuracyOfficialApr 17, 2025Details
AIME (OTIS Mock)Reasoning73.1%accuracyOfficialApr 17, 2025Details
LiveCodeBench ProCoding1288Codeforces EloOfficialApr 17, 2025Details
Humanity's Last ExamReasoning12.1%accuracyOfficialApr 17, 2025Details
LMArenaChat preference1417source-defined ratingOfficialApr 17, 2025Details
WeirdMLCoding40.9%average accuracyOfficialApr 17, 2025Details
MCP AtlasTool use3.4%pass rateOfficialApr 17, 2025Details
Artificial Analysis Intelligence IndexReasoning20.6IndexUnverifiedApr 17, 2025Details
Epoch Capabilities IndexReasoning143.4IndexOfficialApr 17, 2025Details
Aider PolyglotCoding61.9%% correctUnverifiedApr 17, 2025Details
MMLU-ProReasoning83.2%accuracyVerifiedApr 17, 2025Details
GDPvalAgents1071EloOfficialApr 17, 2025Details
LiveCodeBenchCoding49.5%Pass@1UnverifiedApr 17, 2025Details
SciCodeCoding29.1%accuracyUnverifiedApr 17, 2025Details
MMMU (Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark)Multimodal79.7%accuracyUnverifiedApr 17, 2025Details
LongBench v2Reasoning62.1%accuracyOfficialApr 17, 2025Details
Arena-Hard-Auto v2.0Chat preference68.6%% win rateOfficialApr 17, 2025Details
EQ-Bench Creative Writing v3Chat preference1243EloVerifiedApr 17, 2025Details
Design ArenaChat preference1113EloVerifiedApr 17, 2025Details
MASK (Model Alignment between Statements and Knowledge)Other49.13Honesty scoreVerifiedApr 17, 2025Details
MCP-UniverseTool use21.65%Overall Success RateVerifiedApr 17, 2025Details
FACTS GroundingReasoning69.9%Grounding accuracyVerifiedApr 17, 2025Details
EnigmaEvalReasoning2.70%accuracyVerifiedApr 17, 2025Details
ZeroBenchMultimodal3.0% (pass@1)accuracyVerifiedApr 17, 2025Details
FrontierMath Tier 4Reasoning4.2%accuracyOfficialApr 17, 2025Details
Vectara Hallucination LeaderboardOther7.8%Hallucination RateOfficialApr 17, 2025Details