evals.report
BenchmarksLabsCompareRun guidesIn the wild
Google DeepMindGemini

Gemini 3 Flash

Google DeepMind · Gemini. Released Dec 17, 2025.

Gemini 3 Flash is a model from Google DeepMind in the Gemini family, released Dec 17, 2025. evals.report tracks 38 reported Gemini 3 Flash benchmark scores across FrontierMath, DeepSWE, ARC-AGI-1, LMArena, LiveCodeBench Pro, SWE-bench Pro, SWE-bench Verified, Humanity's Last Exam, and 30 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

38 results

Benchmark results 38

Compare this model
BenchmarkCategoryScoreMetricStatusDate
FrontierMathReasoning35.64%accuracyOfficialDec 17, 2025Details
DeepSWECoding5.16%% resolvedOfficialDec 17, 2025Details
ARC-AGI-1Reasoning84.67%accuracyOfficialDec 17, 2025Details
LMArenaChat preference1466source-defined ratingOfficialDec 17, 2025Details
LiveCodeBench ProCoding2316Codeforces EloOfficialDec 17, 2025Details
SWE-bench ProCoding34.63%% resolvedOfficialDec 17, 2025Details
SWE-bench VerifiedCoding75.4%% resolvedOfficialDec 17, 2025Details
Humanity's Last ExamReasoning36.6%accuracyOfficialDec 17, 2025Details
AIME (OTIS Mock)Reasoning92.8%accuracyOfficialDec 17, 2025Details
SimpleQA VerifiedOther67.4%accuracyOfficialDec 17, 2025Details
GPQA DiamondReasoning83.2%accuracyOfficialDec 17, 2025Details
ARC-AGI-2Reasoning33.61%accuracyOfficialDec 17, 2025Details
WeirdMLCoding61.6%average accuracyOfficialDec 17, 2025Details
Artificial Analysis Intelligence IndexReasoning35IndexUnverifiedDec 17, 2025Details
Epoch Capabilities IndexReasoning150.9IndexOfficialDec 17, 2025Details
SWE-rebenchCoding57.6%Resolved rate (pass@1)OfficialDec 17, 2025Details
MMLU-ProReasoning89.0%accuracyVerifiedDec 17, 2025Details
GAIA: A Benchmark for General AI AssistantsAgents35.2%accuracyUnverifiedDec 17, 2025Details
AIME 2026Reasoning95.83%accuracyOfficialDec 17, 2025Details
GDPvalAgents1204EloOfficialDec 17, 2025Details
LiveCodeBenchCoding79.7%Pass@1UnverifiedDec 17, 2025Details
SciCodeCoding50.6%accuracyUnverifiedDec 17, 2025Details
OpenAI-MRCR v2 (Multi-Round Coreference Resolution)Reasoning22.1%accuracy (mean SequenceMatcher similarity)VerifiedDec 17, 2025Details
Global-MMLUReasoning91.4%accuracyUnverifiedDec 17, 2025Details
Video-MMMUMultimodal86.9%accuracyUnverifiedDec 17, 2025Details
WebDev ArenaChat preference1437EloVerifiedDec 17, 2025Details
Search ArenaChat preference1206EloVerifiedDec 17, 2025Details
Design ArenaChat preference1244EloVerifiedDec 17, 2025Details
CharXivMultimodal80.3%accuracyUnverifiedDec 17, 2025Details
ScreenSpot-ProMultimodal69.1%accuracyUnverifiedDec 17, 2025Details
SWE-bench MultilingualCoding72.7%% resolvedOfficialDec 17, 2025Details
ZeroBenchMultimodal13.0% (pass@5)accuracyVerifiedDec 17, 2025Details
MathArena HMMT February 2026Reasoning89.39%accuracyOfficialDec 17, 2025Details
FrontierMath Tier 4Reasoning4.2%accuracyOfficialDec 17, 2025Details
Vectara Hallucination LeaderboardOther13.5%Hallucination RateOfficialDec 17, 2025Details
Vibe Code BenchCoding20.20%Overall accuracyVerifiedDec 17, 2025Details
GSO: Software Optimization Benchmark for SWE-AgentsCoding9.8%Opt@1OfficialDec 17, 2025Details
ProgramBenchCoding0.0%almost-resolved rateOfficialDec 17, 2025Details