Which is better, GPT-5.2 or GPT-5.5?

Across 62 shared benchmarks tracked on Vector Wire, GPT-5.2 leads on 13 and GPT-5.5 leads on 48 — GPT-5.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5.2 vs GPT-5.5

OpenAIvsOpenAI62 shared benchmarks13–48 head-to-head

Benchmark	GPT-5.2	GPT-5.5
AA Agentic Index	60.2	47.4
AA Intelligence	43.3	56.3
AA-LCR	79.3	79
AA-Omniscience	0.3	20.5
aime_2026	98.3	100
apexAgents	23	37.7
arc_agi_1	86.2	95
ARC-AGI-2	52.9	85
arena_elo	1437	1482
arena_text_factuality	1442	1482
arena_vision	1245	1283
Artificial Analysis Coding Index	48.7	74.9
browsecomp	65.8	84.4
coding_arena_elo	1418	1457
critpt	11.6	27.1
Fortress	17.5	16.3
frontiermath_tier_4	18.8	35.4
gdpval	48.3	49.5
GDPval-AA (Elo)	1462	1769
GPQA Diamond	92.4	93.6
HealthBench Consensus length-adjusted	94.4	95.6
HealthBench Hard length-adjusted	34.3	31.5
HealthBench Professional length-adjusted	45.9	51.8
healthbench_length_adjusted	56.8	56.5
HLE	45.5	52.2
HLE (with tools)	45.5	52.2
hmmt_feb_2026	97	98.5
hmmt_nov_2025	99.2	96.5
IFBench	75.4	75.9
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
livebench	74.8	80.7
livebench_agentic_coding	50.3	54
livebench_coding	76.1	82.2
livebench_data_analysis	78.2	81.6
livebench_instruction_following	61.8	70.7
livebench_language	79.8	87.4
livebench_math	93.2	95.9
livebench_reasoning	83.2	89.7
matharena_visual_math_overall	86.5	94.9
MCP Atlas	68	82.8
MMMU-Pro	79.5	83.2
OmniScience Accuracy	44.3	58
OmniScience Non-Hallucination	38.4	12
scicode	52.1	56.1
simplebench	45.8	69
simpleqa_verified	38.9	63.1
SWE-bench Pro	55.6	58.6
SWE-bench Verified	80	80.6
TauBench V3 - Banking	11.1	39
Terminal-Bench 2.0	54	82.7
Terminal-Bench Hard	62.2	60.6
tool_decathlon	46.3	55.6
toolathlon	46.3	55.6
vectara_answer_rate	100	100
vectara_avg_summary_length	186.3	129.6
vectara_factual_consistency	89.2	90.7
vectara_hallucination_rate ↓	10.8	9.3
τ²-Bench Telecom (AA run)	98.7	93.9
τ³-Bench	12	31.3

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.