Which is better, GPT-5 or GPT-5.1?

Across 53 shared benchmarks tracked on Vector Wire, GPT-5 leads on 24 and GPT-5.1 leads on 26 — GPT-5.1 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-5.1

OpenAIvsOpenAI53 shared benchmarks24–26 head-to-head

Benchmark	GPT-5	GPT-5.1
AA Agentic Index	49.7	32.2
AA Intelligence	35.3	37.5
AA-LCR	76.3	76.7
AA-Omniscience	-8.7	5.4
AIME 2025	95	94.2
arc_agi_1	56.2	72.8
ARC-AGI-2	7.5	17.6
arena_vision	1212	1250
Artificial Analysis Coding Index	38.9	49.4
browsecomp	54.9	50.8
coding_arena_elo	1419	1421
critpt	5.7	4.9
Fortress	17	25.7
frontiermath_tier_4	12.5	12.5
gdpval	32.5	25
GPQA Diamond	87.3	88.1
HealthBench Consensus length-adjusted	95.6	95
HealthBench Hard length-adjusted	34.7	25.4
HealthBench Professional length-adjusted	46.2	39.6
healthbench_length_adjusted	57.7	50.9
HLE	28.5	28.5
HLE (with tools)	35.2	42.7
hmmt_feb_2025	88.3	96.3
hmmt_nov_2025	89.2	91.7
IFBench	73.1	72.9
LiveCodeBench v6	87	87
matharena_visual_math_overall	78.8	76.9
MMLU-Pro	87.5	87
MMMU	84.2	85.4
MMMU-Pro	78.4	75.5
multichallenge	71.1	63.4
MultiNRC	52.1	49
OmniScience Accuracy	40.3	37.7
OmniScience Non-Hallucination	21	48.1
scicode	43	43.3
SEAL VISTA	49.7	43.8
simplebench	56.7	53.2
simpleqa_verified	50.6	48.9
swe_bench_bash	65	66
SWE-bench Verified	74.9	76.3
Tau2 airline	62.6	67
Tau2 retail	81.1	77.9
TauBench V3 - Banking	22.1	15.9
Terminal-Bench 2.0	35.2	47.6
Terminal-Bench 2.1	35.2	52.4
Terminal-Bench Hard	37.9	45.5
vectara_answer_rate	99.9	100
vectara_avg_summary_length	162.7	254.4
vectara_factual_consistency	85.3	89.1
vectara_hallucination_rate ↓	14.7	12.1
τ²-Bench	85	82.7
τ²-Bench Telecom (AA run)	86.5	81.9
τ³-Bench	19.6	14

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.