evals.report
BenchmarksLabsCompareRun guidesIn the wild
OpenAIGPT

GPT-5.4

OpenAI · GPT. Released Mar 5, 2026.

GPT-5.4 is a model from OpenAI in the GPT family, released Mar 5, 2026. evals.report tracks 46 reported GPT-5.4 benchmark scores across FrontierMath, DeepSWE, ARC-AGI-1, ARC-AGI-3, ARC-AGI-2, LMArena, LiveBench, SWE-bench Pro, and 38 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

46 results

Benchmark results 46

Compare this model
BenchmarkCategoryScoreMetricStatusDate
FrontierMathReasoning47.6%accuracyOfficialMar 5, 2026Details
DeepSWECoding55.53%% resolvedOfficialMar 5, 2026Details
ARC-AGI-1Reasoning93.67%accuracyOfficialMar 5, 2026Details
ARC-AGI-3Reasoning0.21%accuracyOfficialMar 5, 2026Details
ARC-AGI-2Reasoning73.95%accuracyOfficialMar 5, 2026Details
LMArenaChat preference1472source-defined ratingOfficialMar 5, 2026Details
LiveBenchReasoning80.28%scoreOfficialMar 5, 2026Details
SWE-bench ProCoding59.10%% resolvedOfficialMar 5, 2026Details
GPQA DiamondReasoning93.3%accuracyOfficialMar 5, 2026Details
SWE-bench VerifiedCoding76.9%% resolvedOfficialMar 5, 2026Details
Humanity's Last ExamReasoning40.28%accuracyOfficialMar 5, 2026Details
MMMU-ProMultimodal82.1%accuracyOfficialMar 5, 2026Details
AIME (OTIS Mock)Reasoning95.3%accuracyOfficialMar 5, 2026Details
SimpleQA VerifiedOther44.8%accuracyOfficialMar 5, 2026Details
GBA EvalCoding31.6%overall scoreOfficialMar 5, 2026Details
WeirdMLCoding77.7%average accuracyOfficialMar 5, 2026Details
PostTrainBenchAgents20.23%weighted average scoreOfficialMar 5, 2026Details
Artificial Analysis Intelligence IndexReasoning56.8IndexUnverifiedMar 5, 2026Details
Epoch Capabilities IndexReasoning156.1IndexOfficialMar 5, 2026Details
OSWorldAgents75.0%task success rateUnverifiedMar 5, 2026Details
GAIA: A Benchmark for General AI AssistantsAgents48.2%accuracyUnverifiedMar 5, 2026Details
τ²-bench (Telecom)Tool use87.1%pass^1OfficialMar 5, 2026Details
AIME 2026Reasoning99.17%accuracyOfficialMar 5, 2026Details
GDPvalAgents1674EloOfficialMar 5, 2026Details
METR Task-Completion Time HorizonsAgents341.7 min50% time horizonOfficialMar 5, 2026Details
SciCodeCoding56.6%accuracyUnverifiedMar 5, 2026Details
AA-Omniscience: Knowledge and Hallucination BenchmarkReasoning6AA-Omniscience IndexOfficialMar 5, 2026Details
IFBenchReasoning73.9%accuracyOfficialMar 5, 2026Details
WebDev ArenaChat preference1388EloVerifiedMar 5, 2026Details
Search ArenaChat preference1199EloVerifiedMar 5, 2026Details
EQ-Bench Creative Writing v3Chat preference2003EloVerifiedMar 5, 2026Details
Design ArenaChat preference1264EloVerifiedMar 5, 2026Details
MASK (Model Alignment between Statements and Knowledge)Other89.67Honesty scoreVerifiedMar 5, 2026Details
ScreenSpot-ProMultimodal85.4%accuracyUnverifiedMar 5, 2026Details
EnigmaEvalReasoning15.96%accuracyVerifiedMar 5, 2026Details
ZeroBenchMultimodal23.0% (pass@5)accuracyVerifiedMar 5, 2026Details
MathArena HMMT February 2026Reasoning97.73%accuracyOfficialMar 5, 2026Details
FrontierMath Tier 4Reasoning27.1%accuracyOfficialMar 5, 2026Details
Vectara Hallucination LeaderboardOther7.0%Hallucination RateOfficialMar 5, 2026Details
Vibe Code BenchCoding67.42%Overall accuracyOfficialMar 5, 2026Details
Online-Mind2WebAgents92.8%Task success rateVerifiedMar 5, 2026Details
GSO: Software Optimization Benchmark for SWE-AgentsCoding31.37%Opt@1OfficialMar 5, 2026Details
MultiNRCReasoning58.29%accuracyOfficialMar 5, 2026Details
Terminal-Bench 2.0Agents75.1%task successVerifiedMar 5, 2026Details
FrontierSWEAgents54%dominance scoreOfficialMar 5, 2026Details
ProgramBenchCoding0.0%almost-resolved rateOfficialMar 5, 2026Details