evals.report
BenchmarksLabsCompareRun guidesIn the wild
Google DeepMindGemini

Gemini 3.1 Pro Preview

Google DeepMind · Gemini. Released Feb 19, 2026.

Gemini 3.1 Pro Preview is a model from Google DeepMind in the Gemini family, released Feb 19, 2026. evals.report tracks 50 reported Gemini 3.1 Pro Preview benchmark scores across FrontierMath, DeepSWE, ARC-AGI-1, ARC-AGI-3, ARC-AGI-2, LMArena, LiveBench, LiveCodeBench Pro, and 42 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

50 results

Benchmark results 50

Compare this model
BenchmarkCategoryScoreMetricStatusDate
FrontierMathReasoning36.9%accuracyOfficialFeb 19, 2026Details
DeepSWECoding9.88%% resolvedOfficialFeb 19, 2026Details
ARC-AGI-1Reasoning98%accuracyOfficialFeb 19, 2026Details
ARC-AGI-3Reasoning0.42%accuracyOfficialFeb 19, 2026Details
ARC-AGI-2Reasoning77.08%accuracyOfficialFeb 19, 2026Details
LMArenaChat preference1481source-defined ratingOfficialFeb 19, 2026Details
LiveBenchReasoning79.93%scoreOfficialFeb 19, 2026Details
LiveCodeBench ProCoding2887Codeforces EloOfficialFeb 19, 2026Details
SWE-bench ProCoding46.10%% resolvedOfficialFeb 19, 2026Details
GPQA DiamondReasoning94.1%accuracyOfficialFeb 19, 2026Details
SWE-bench VerifiedCoding75.6%% resolvedOfficialFeb 19, 2026Details
Humanity's Last ExamReasoning45.9%accuracyOfficialFeb 19, 2026Details
MMMU-ProMultimodal80.5%accuracyOfficialFeb 19, 2026Details
AIME (OTIS Mock)Reasoning95.6%accuracyOfficialFeb 19, 2026Details
SimpleQA VerifiedOther77.3%accuracyOfficialFeb 19, 2026Details
GBA EvalCoding0.8%overall scoreOfficialFeb 19, 2026Details
WeirdMLCoding72.1%average accuracyOfficialFeb 19, 2026Details
PostTrainBenchAgents21.59%weighted average scoreOfficialFeb 19, 2026Details
Artificial Analysis Intelligence IndexReasoning57IndexOfficialFeb 19, 2026Details
Epoch Capabilities IndexReasoning156.6IndexOfficialFeb 19, 2026Details
MMLU-ProReasoning90.99%accuracyVerifiedFeb 19, 2026Details
OSWorldAgents76.2%task success rateVerifiedFeb 19, 2026Details
GAIA: A Benchmark for General AI AssistantsAgents46.1%accuracyUnverifiedFeb 19, 2026Details
τ²-bench (Telecom)Tool use95.6%pass^1OfficialFeb 19, 2026Details
AIME 2026Reasoning98.33%accuracyOfficialFeb 19, 2026Details
GDPvalAgents1314EloOfficialFeb 19, 2026Details
METR Task-Completion Time HorizonsAgents384.1 min50% time horizonOfficialFeb 19, 2026Details
SciCodeCoding58.9%accuracyUnverifiedFeb 19, 2026Details
AA-Omniscience: Knowledge and Hallucination BenchmarkReasoning33AA-Omniscience IndexOfficialFeb 19, 2026Details
IFBenchReasoning77.1%accuracyOfficialFeb 19, 2026Details
MultiChallengeReasoning71.37%accuracyVerifiedFeb 19, 2026Details
OpenAI-MRCR v2 (Multi-Round Coreference Resolution)Reasoning26.3%accuracy (mean SequenceMatcher similarity)UnverifiedFeb 19, 2026Details
Global-MMLUReasoning93.2%accuracyUnverifiedFeb 19, 2026Details
WebDev ArenaChat preference1448EloVerifiedFeb 19, 2026Details
Search ArenaChat preference1216EloVerifiedFeb 19, 2026Details
EQ-Bench Creative Writing v3Chat preference1479EloVerifiedFeb 19, 2026Details
Design ArenaChat preference1282EloVerifiedFeb 19, 2026Details
MASK (Model Alignment between Statements and Knowledge)Other42.40Honesty scoreVerifiedFeb 19, 2026Details
ScreenSpot-ProMultimodal84.4%accuracyUnverifiedFeb 19, 2026Details
EnigmaEvalReasoning19.76%accuracyVerifiedFeb 19, 2026Details
ZeroBenchMultimodal19.0% (pass@5)accuracyVerifiedFeb 19, 2026Details
MathArena HMMT February 2026Reasoning94.70%accuracyOfficialFeb 19, 2026Details
FrontierMath Tier 4Reasoning16.7%accuracyOfficialFeb 19, 2026Details
Vectara Hallucination LeaderboardOther10.4%Hallucination RateOfficialFeb 19, 2026Details
Vibe Code BenchCoding32.03%Overall accuracyVerifiedFeb 19, 2026Details
GSO: Software Optimization Benchmark for SWE-AgentsCoding22.55%Opt@1OfficialFeb 19, 2026Details
MultiNRCReasoning64.74%accuracyOfficialFeb 19, 2026Details
FrontierSWEAgents40%dominance scoreOfficialFeb 19, 2026Details
ProgramBenchCoding0.0%almost-resolved rateOfficialFeb 19, 2026Details
FrontierCodeCoding4.7%weighted score (Diamond)OfficialFeb 19, 2026Details