Which is better, Claude Sonnet 4.5 or GPT-5?

Across 83 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 29 and GPT-5 leads on 54 — GPT-5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Sonnet 4.5 vs GPT-5

AnthropicvsOpenAI83 shared benchmarks29–54 head-to-head

Benchmark	Claude Sonnet 4.5	GPT-5
AA Agentic Index	50.6	49.7
AA Intelligence	37.4	35.3
AA-LCR	68.3	76.3
AA-Omniscience	-0.1	-8.7
AIME 2025	88	95
AIME25 no tools	88	94.6
AIME25 w/ python	100	99.6
AIR-Bench 2024	89.8	87.7
arc_agi_1	63.7	56.2
ARC-AGI-2	13.6	7.5
Arena-Hard (Creative Writing)	76.7	92.2
Arena-Hard (Hard Prompt)	63.3	71.9
ArtifactsBench	61.5	73
Artificial Analysis Coding Index	52.1	38.9
browsecomp	24.1	54.9
BrowseComp w/ tools	24.1	54.9
browsecomp_zh	42.4	65
BrowseComp-ZH w/ tools	42.4	63
coding_arena_elo	1392	1419
critpt	1.1	5.7
FinSearchComp-global	60.8	63.9
FinSearchComp-T3	44	48.5
FinSearchComp-T3 w/ tools	44	48.5
Fortress	17	17
Frames	85	86
Frames w/ tools	85	86
frontiermath_tier_4	4.2	12.5
GAIA (text only)	71.2	76.4
gdpval	40.4	32.5
GPQA (unspecified)	83.4	85.7
GPQA Diamond	83.4	87.3
HealthBench	44.2	67.2
HealthBench no tools	44.2	67.2
HLE	19.8	28.5
HLE (with tools)	32	35.2
HMMT 2025	74.6	93.3
hmmt_feb_2025	79.2	88.3
hmmt_nov_2025	81.7	89.2
HMMT25 no tools	74.6	93.3
HMMT25 w/ python	88.8	96.7
IFBench	57.3	73.1
imo_answer_bench	65.9	76
LiveCodeBench	71	85
LiveCodeBench v6	64	87
LiveCodeBenchV6 no tools	64	87
Longform Writing eval (Kimi K2 Thinking system card)	79.8	71.4
matharena_visual_math_overall	75.8	78.8
mmlu_redux	95.6	95.3
MMLU-Pro	88.2	87.5
MMMU	79.6	84.2
MMMU-Pro	68.7	78.4
Multi-SWE-Bench	44.3	39.3
multichallenge	55.3	71.1
MultiNRC	35.8	52.1
OJ-Bench (cpp)	30.4	56.2
OJ-Bench (cpp) no tools	30.4	56.2
OmniScience Accuracy	32.9	40.3
OmniScience Non-Hallucination	50.9	21
scicode	45	43
SEAL VISTA	48.8	49.7
Seal-0	53.4	51.4
Seal-0 w/ tools	53.4	51.4
simplebench	54.3	56.7
simpleqa_verified	23.6	50.6
swe_bench_bash	71.4	65
SWE-bench Multilingual	68	55.3
SWE-bench Pro	43.6	41.8
SWE-bench Verified	82	74.9
TauBench V3 - Banking	24.5	22.1
Terminal-Bench	51	43.8
Terminal-Bench 2.0	50	35.2
Terminal-Bench 2.1	55.8	35.2
Terminal-Bench Hard	35.6	37.9
Terminal-Bench w/ simulated tools (JSON)	51	43.8
vectara_answer_rate	95.6	99.9
vectara_avg_summary_length	127.8	162.7
vectara_factual_consistency	88	85.3
vectara_hallucination_rate ↓	12	14.7
VisualToolBench	6.2	18.7
xbench_deepsearch	66	77.8
τ²-Bench	98	85
τ²-Bench Telecom (AA run)	78.1	86.5
τ³-Bench	19	19.6

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.