Which is better, GPT-5.1 or GPT-5.2?

Across 56 shared benchmarks tracked on Vector Wire, GPT-5.1 leads on 17 and GPT-5.2 leads on 37 — 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-5.1 vs GPT-5.2

OpenAIvsOpenAI56 shared benchmarks17–37 head-to-head

Benchmark	GPT-5.1	GPT-5.2
AA Agentic Index	32.2	60.2
AA Intelligence	37.5	43.3
AA-LCR	76.7	79.3
AA-Omniscience	5.4	0.3
AIME 2025	94.2	100
arc_agi_1	72.8	86.2
ARC-AGI-2	17.6	52.9
arena_elo	1439	1437
arena_text_factuality	1445	1442
arena_vision	1250	1245
Artificial Analysis Coding Index	49.4	48.7
browsecomp	50.8	65.8
coding_arena_elo	1421	1418
critpt	4.9	11.6
Fortress	25.7	17.5
frontiermath_tier_4	12.5	18.8
gdpval	25	48.3
GPQA Diamond	88.1	92.4
HealthBench Consensus length-adjusted	95	94.4
HealthBench Hard length-adjusted	25.4	34.3
HealthBench Professional length-adjusted	39.6	45.9
healthbench_length_adjusted	50.9	56.8
HLE	28.5	45.5
HLE (with tools)	42.7	45.5
hmmt_feb_2025	96.3	100
hmmt_nov_2025	91.7	99.2
IFBench	72.9	75.4
Image input eval - extremism (not_unsafe)	1	1
Image input eval - harms-erotic (not_unsafe)	1	1
Image input eval - hate (not_unsafe)	1	1
Image input eval - self-harm (not_unsafe)	1	1
matharena_visual_math_overall	76.9	86.5
MCP Atlas	50.1	68
MMLU-Pro	87	87
mmmlu	91	89.6
MMMU-Pro	75.5	79.5
MultiNRC	49	42.2
OmniScience Accuracy	37.7	44.3
OmniScience Non-Hallucination	48.1	38.4
scicode	43.3	52.1
SEAL VISTA	43.8	46.6
simplebench	53.2	45.8
simpleqa_verified	48.9	38.9
swe_bench_bash	66	72.8
SWE-bench Verified	76.3	80
Tau2 retail	77.9	82
TauBench V3 - Banking	15.9	11.1
Terminal-Bench 2.0	47.6	54
Terminal-Bench Hard	45.5	62.2
vectara_answer_rate	100	100
vectara_avg_summary_length	254.4	186.3
vectara_factual_consistency	89.1	89.2
vectara_hallucination_rate ↓	12.1	10.8
τ²-Bench	82.7	85.5
τ²-Bench Telecom (AA run)	81.9	98.7
τ³-Bench	14	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.