Which is better, GPT-5.5 or Opus 5?

Across 47 shared benchmarks tracked on Vector Wire, GPT-5.5 leads on 9 and Opus 5 leads on 38 — Opus 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-5.5 vs Opus 5

OpenAIvsAnthropic47 shared benchmarks9–38 head-to-head

Benchmark	GPT-5.5	Opus 5
AA Agentic Index	47.4	59.2
AA Intelligence	56.3	63.1
AA-LCR	79	78.7
AA-Omniscience	20.5	37.1
arc_agi_1	95	97.5
arc_agi_3	0.4	30.2
ARC-AGI-2	85	90.4
arena_elo	1482	1487
arena_text_factuality	1482	1481
Artificial Analysis Coding Index	74.9	78
browsecomp	84.4	90.8
coding_arena_elo	1457	1662
critpt	27.1	29.1
DeepSWE 1.1	67	68.8
Finance Agent v2	51.8	58.6
GDP (Surge AI)	16.7	85.5
gdpval	49.5	67.2
GDPval-AA (Elo)	1769	1827
GDPval-AA v2	1509	1861
GPQA Diamond	93.6	93.7
HealthBench	56.5	67.1
healthbench_length_adjusted	56.5	57.8
healthbench_professional	51.8	59.8
HiL-Bench	39.7	57
HLE	52.2	64.7
HLE (with tools)	52.2	64.7
Legal Agent Benchmark	2.1	23
livebench_agentic_coding	54	65.2
livebench_coding	82.2	81.5
livebench_data_analysis	81.6	74.5
livebench_instruction_following	70.7	63.8
livebench_language	87.4	88.7
livebench_math	95.9	95.7
livebench_reasoning	89.7	91.2
MCP Atlas	82.8	85.8
MMMU-Pro	83.2	84.7
officeqa_pro	60.9	66.9
OmniScience Accuracy	58	60.9
OmniScience Non-Hallucination	12	40.5
OSWorld-Verified	78.7	4
scicode	56.1	55.7
simplebench	69	80.6
simpleqa_verified	63.1	56.7
SWE-bench Pro	58.6	79.2
SWE-bench Verified	80.6	96
TauBench V3 - Banking	39	44.7
Terminal-Bench 2.1	84.3	89.1

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.