Which is better, GPT-5.4 or GPT-5.5?

Across 73 shared benchmarks tracked on Vector Wire, GPT-5.4 leads on 14 and GPT-5.5 leads on 59 — 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-5.4 vs GPT-5.5

OpenAIvsOpenAI73 shared benchmarks14–59 head-to-head

Benchmark	GPT-5.4	GPT-5.5
AA Agentic Index	58.2	47.4
AA Intelligence	53.1	56.3
AA-LCR	77.7	79
AA-Omniscience	5.8	20.5
aime_2026	99.2	100
apexAgents	33.3	37.7
arc_agi_1	93.7	95
arc_agi_3	0.2	0.4
ARC-AGI-2	74	85
arena_elo	1477	1482
arena_text_factuality	1476	1482
arena_vision	1277	1283
Artificial Analysis Coding Index	71.1	74.9
browsecomp	82.7	84.4
coding_arena_elo	1457	1457
critpt	23.4	27.1
cybergym	66.3	81.8
DeepSWE 1.1	52	67
DNA sequence design for transcription factor binding	12.8	13.8
finance_agent	57.2	60
frontiermath_tier_4	27.1	35.4
gdpval	50.1	49.5
GDPval-AA (Elo)	1674	1769
GPQA Diamond	93	93.6
Hard-negative protein binding prediction	3.5	1.5
HealthBench Consensus length-adjusted	96.3	95.6
HealthBench Hard length-adjusted	29.1	31.5
HealthBench Professional length-adjusted	48.1	51.8
healthbench_length_adjusted	54	56.5
HiL-Bench	9.7	39.7
HLE	43.7	52.2
HLE (with tools)	52.1	52.2
hmmt_feb_2026	97.7	98.5
hmmt_nov_2025	95.8	96.5
IFBench	73.9	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
itbenchSre	34.5	45.8
Legal Agent Benchmark	0.4	2.1
livebench	80.3	80.7
livebench_agentic_coding	53.8	54
livebench_coding	77.5	82.2
livebench_data_analysis	79.3	81.6
livebench_instruction_following	70.2	70.7
livebench_language	82.6	87.4
livebench_math	94.2	95.9
livebench_reasoning	88.1	89.7
matharena_visual_math_overall	92.5	94.9
MCP Atlas	70.6	82.8
MMMU-Pro	81.2	83.2
nl2repo	41.3	50.7
officeqa_pro	51.1	60.9
OmniScience Accuracy	50.9	58
OmniScience Non-Hallucination	17.4	12
OSWorld-Verified	75	78.7
scicode	56.6	56.1
simpleqa_verified	45.3	63.1
SWE-bench Pro	59.1	58.6
TauBench V3 - Banking	39.6	39
Terminal-Bench 2.0	75.1	82.7
Terminal-Bench 2.1	78.3	84.3
Terminal-Bench Hard	57.6	60.6
tool_decathlon	54.6	55.6
toolathlon	54.6	55.6
usamo_2026	95.2	98.2
vectara_answer_rate	99.9	100
vectara_avg_summary_length	81.7	129.6
vectara_factual_consistency	93	90.7
vectara_hallucination_rate ↓	7	9.3
τ²-Bench Telecom (AA run)	87.1	93.9
τ³-Bench	72.9	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.