Which is better, GPT-5 or GPT-5.4?

Across 45 shared benchmarks tracked on Vector Wire, GPT-5 leads on 5 and GPT-5.4 leads on 38 — GPT-5.4 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5 vs GPT-5.4

OpenAIvsOpenAI45 shared benchmarks5–38 head-to-head

Benchmark	GPT-5	GPT-5.4
AA Agentic Index	49.7	58.2
AA Intelligence	35.3	53.1
AA-LCR	76.3	77.7
AA-Omniscience	-8.7	5.8
arc_agi_1	56.2	93.7
ARC-AGI-2	7.5	74
arena_vision	1212	1277
Artificial Analysis Coding Index	38.9	71.1
browsecomp	54.9	82.7
coding_arena_elo	1419	1457
critpt	5.7	23.4
frontiermath_tier_4	12.5	27.1
gdpval	32.5	50.1
GPQA Diamond	87.3	93
HealthBench Consensus length-adjusted	95.6	96.3
HealthBench Hard length-adjusted	34.7	29.1
HealthBench Professional length-adjusted	46.2	48.1
healthbench_length_adjusted	57.7	54
HLE	28.5	43.7
HLE (with tools)	35.2	52.1
hmmt_nov_2025	89.2	95.8
IFBench	73.1	73.9
imo_answer_bench	76	91.4
matharena_visual_math_overall	78.8	92.5
MMLU-Pro	87.5	87.5
MMMU-Pro	78.4	81.2
MultiNRC	52.1	58.3
OmniScience Accuracy	40.3	50.9
OmniScience Non-Hallucination	21	17.4
scicode	43	56.6
SEAL VISTA	49.7	50.9
simpleqa_verified	50.6	45.3
SWE-bench Multilingual	55.3	71.7
SWE-bench Pro	41.8	59.1
TauBench V3 - Banking	22.1	39.6
Terminal-Bench 2.0	35.2	75.1
Terminal-Bench 2.1	35.2	78.3
Terminal-Bench Hard	37.9	57.6
vectara_answer_rate	99.9	99.9
vectara_avg_summary_length	162.7	81.7
vectara_factual_consistency	85.3	93
vectara_hallucination_rate ↓	14.7	7
VisualToolBench	18.7	29.2
τ²-Bench Telecom (AA run)	86.5	87.1
τ³-Bench	19.6	72.9

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.