Which is better, Claude Opus 4.8 or GPT-5.4?

Across 72 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 43 and GPT-5.4 leads on 29 — 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-5.4

AnthropicvsOpenAI72 shared benchmarks43–29 head-to-head

Benchmark	Claude Opus 4.8	GPT-5.4
AA Agentic Index	49.4	58.2
AA Intelligence	57.3	53.1
AA-LCR	73	77.7
AA-Omniscience	28.8	5.8
AgentWorldBench - Android	61.5	60
AgentWorldBench - MCP	54.9	70.1
AgentWorldBench - OS	66.6	68.6
AgentWorldBench - Overall	56.6	58.3
AgentWorldBench - Search	35.1	37.3
AgentWorldBench - SWE	64.1	66.3
AgentWorldBench - Terminal	59.2	53.7
AgentWorldBench - Web	54.7	51.8
aime_2026	100	99.2
arc_agi_1	92.5	93.7
arc_agi_3	1.5	0.2
ARC-AGI-2	72.1	74
arena_elo	1482	1477
arena_text_factuality	1464	1476
arena_vision	1280	1277
Artificial Analysis Coding Index	74.3	71.1
baby_vision_with_python	81.2	80.2
browsecomp	88.5	82.7
charxiv_rq	80.5	82.8
coding_arena_elo	1539	1457
critpt	20.9	23.4
cybergym	83.1	66.3
deepsearchqa_f1	93.1	78.6
DeepSWE 1.1	59	52
finance_agent	53.9	57.2
frontiermath_tier_4	31.3	27.1
gdpval	54.2	50.1
GDPval-AA (Elo)	1890	1674
GPQA Diamond	93.6	93
HiL-Bench	35.3	9.7
HLE	57.9	43.7
HLE (with tools)	57.9	52.1
hmmt_feb_2026	96.7	97.7
hmmt_nov_2025	96.5	95.8
IFBench	62.2	73.9
imo_answer_bench	83.5	91.4
Legal Agent Benchmark	10	0.4
livebench	77.2	80.3
livebench_agentic_coding	50.5	53.8
livebench_coding	81.8	77.5
livebench_data_analysis	66	79.3
livebench_instruction_following	72	70.2
livebench_language	79.7	82.6
livebench_math	94.3	94.2
livebench_reasoning	89.2	88.1
matharena_visual_math_overall	81.6	92.5
mathvision	86.7	92
MCP Atlas	83.6	70.6
MMMU-Pro	78.9	81.2
nl2repo	69.7	41.3
officeqa	77.6	68.1
officeqa_pro	66.2	51.1
OmniScience Accuracy	48.8	50.9
OmniScience Non-Hallucination	60.7	17.4
OSWorld-Verified	83.4	75
scicode	53.5	56.6
simpleqa_verified	39.5	45.3
SWE-bench Multilingual	84.4	71.7
SWE-bench Pro	69.2	59.1
TauBench V3 - Banking	34.2	39.6
Terminal-Bench 2.0	74.6	75.1
Terminal-Bench 2.1	85	78.3
Terminal-Bench Hard	58.3	57.6
tool_decathlon	59.9	54.6
toolathlon	59.9	54.6
usamo_2026	96.7	95.2
τ²-Bench Telecom (AA run)	94.4	87.1
τ³-Bench	27.6	72.9

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.