Which is better, Claude Opus 4.8 or GPT-5.5?

Across 85 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 45 and GPT-5.5 leads on 36 — Claude Opus 4.8 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Opus 4.8 vs GPT-5.5

AnthropicvsOpenAI85 shared benchmarks45–36 head-to-head

Benchmark	Claude Opus 4.8	GPT-5.5
AA Agentic Index	49.4	47.4
AA Intelligence	57.3	56.3
AA-LCR	73	79
AA-Omniscience	28.8	20.5
Agents' Last Exam	27	26.6
aime_2026	100	100
AIRS-Bench	84	86
apex_agents	39.4	38.5
arc_agi_1	92.5	95
arc_agi_3	1.5	0.4
ARC-AGI-2	72.1	85
arena_elo	1482	1482
arena_text_factuality	1464	1482
arena_vision	1280	1283
Artificial Analysis Coding Index	74.3	74.9
browsecomp	88.5	84.4
BrowseComp (single-agent)	88.5	84.4
coding_arena_elo	1539	1457
critpt	20.9	27.1
cybergym	83.1	81.8
DeepSWE	59	70
DeepSWE 1.1	59	67
Finance Agent v2	53.9	51.8
finance_agent	53.9	60
Fortress	18.2	16.3
frontiermath_tier_4	31.3	35.4
GDP (Surge AI)	84.8	16.7
gdpval	54.2	49.5
GDPval-AA (Elo)	1890	1769
GDPval-AA v2	1593	1509
GDPval-AA v2 (Elo)	1593	1491
GPQA Diamond	93.6	93.6
GraphWalks — Parents task, 256K-token context subset	99.3	90.1
GraphWalks BFS 1M subset	68.1	45.4
GraphWalks BFS 256K	85.9	73.7
GraphWalks Parents 1M subset	83.3	58.5
HealthBench	59.3	56.5
healthbench_professional	57.4	51.8
HiL-Bench	35.3	39.7
HLE	57.9	52.2
HLE (with tools)	57.9	52.2
HLE Calibration	26.5	44.2
hmmt_feb_2026	96.7	98.5
hmmt_nov_2025	96.5	96.5
IFBench	62.2	75.9
JobBench	48.4	38.3
Legal Agent Benchmark	10	2.1
livebench	77.2	80.7
livebench_agentic_coding	50.5	54
livebench_coding	81.8	82.2
livebench_data_analysis	66	81.6
livebench_instruction_following	72	70.7
livebench_language	79.7	87.4
livebench_math	94.3	95.9
livebench_reasoning	89.2	89.7
matharena_visual_math_overall	81.6	94.9
MCP Atlas	83.6	82.8
MCP Mark Verified	76.4	92.9
MLS Bench Lite	42.8	35.5
MMMU-Pro	78.9	83.2
nl2repo	69.7	50.7
officeqa_pro	66.2	60.9
OmniScience Accuracy	48.8	58
OmniScience Non-Hallucination	60.7	12
OR-Bench (FRR)	3.3	6.7
OSWorld-Verified	83.4	78.7
PostTrainBench	37.2	28.4
Program Bench	71.9	70.8
ResearchRubrics	73.5	64
scicode	53.5	56.1
simplebench	64.8	69
simpleqa_verified	39.5	63.1
SpreadsheetBench 2	31.6	29.1
SWE-bench Pro	69.2	58.6
SWE-bench Verified	88.6	80.6
SWE-Marathon	40	14
TauBench V3 - Banking	34.2	39
Terminal-Bench 2.0	74.6	82.7
Terminal-Bench 2.1	85	84.3
Terminal-Bench Hard	58.3	60.6
tool_decathlon	59.9	55.6
toolathlon	59.9	55.6
usamo_2026	96.7	98.2
τ²-Bench Telecom (AA run)	94.4	93.9
τ³-Bench	27.6	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.