Which is better, GPT-5 or GPT-5.2?

Across 66 shared benchmarks tracked on Vector Wire, GPT-5 leads on 16 and GPT-5.2 leads on 50 — GPT-5.2 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-5.2

OpenAIvsOpenAI66 shared benchmarks16–50 head-to-head

Benchmark	GPT-5	GPT-5.2
AA Agentic Index	49.7	60.2
AA Intelligence	35.3	43.3
AA-LCR	76.3	79.3
AA-Omniscience	-8.7	0.3
AIME 2025	95	100
AIME25 no tools	94.6	98
arc_agi_1	56.2	86.2
ARC-AGI-2	7.5	52.9
arena_vision	1212	1245
Artificial Analysis Coding Index	38.9	48.7
browsecomp	54.9	65.8
browsecomp_zh	65	76.1
coding_arena_elo	1419	1418
critpt	5.7	11.6
Fortress	17	17.5
frontiermath_tier_4	12.5	18.8
gaia_level1	78.5	92.5
gaia_level2	64.8	74.8
gaia_level3	46.9	57.1
gaia_overall	66.1	77.4
gdpval	32.5	48.3
GPQA (unspecified)	85.7	92.4
GPQA Diamond	87.3	92.4
HealthBench Consensus length-adjusted	95.6	94.4
HealthBench Hard length-adjusted	34.7	34.3
HealthBench Professional length-adjusted	46.2	45.9
healthbench_length_adjusted	57.7	56.8
HLE	28.5	45.5
HLE (with tools)	35.2	45.5
HMMT 2025	93.3	99.4
hmmt_feb_2025	88.3	100
hmmt_nov_2025	89.2	99.2
IFBench	73.1	75.4
imo_answer_bench	76	86.3
LiveCodeBench	85	89
matharena_visual_math_overall	78.8	86.5
metr_hcast	68.4	77
metr_re_bench	40	35
metr_swaa	100	99.6
MMLU-Pro	87.5	87
MMMU-Pro	78.4	79.5
MultiNRC	52.1	42.2
ocrbench_v2	55.5	50.5
OmniScience Accuracy	40.3	44.3
OmniScience Non-Hallucination	21	38.4
scicode	43	52.1
SEAL VISTA	49.7	46.6
Seal-0	51.4	45
simplebench	56.7	45.8
simpleqa_verified	50.6	38.9
swe_bench_bash	65	72.8
SWE-bench Multilingual	55.3	72
SWE-bench Pro	41.8	55.6
SWE-bench Verified	74.9	80
Tau2 retail	81.1	82
TauBench V3 - Banking	22.1	11.1
Terminal-Bench 2.0	35.2	54
Terminal-Bench Hard	37.9	62.2
vectara_answer_rate	99.9	100
vectara_avg_summary_length	162.7	186.3
vectara_factual_consistency	85.3	89.2
vectara_hallucination_rate ↓	14.7	10.8
VideoMMMU	84.6	85.9
τ²-Bench	85	85.5
τ²-Bench Telecom (AA run)	86.5	98.7
τ³-Bench	19.6	12

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.