Which is better, Claude Sonnet 4.5 or GPT-5.1?

Across 51 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 19 and GPT-5.1 leads on 32 — GPT-5.1 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-5.1

AnthropicvsOpenAI51 shared benchmarks19–32 head-to-head

Benchmark	Claude Sonnet 4.5	GPT-5.1
AA Agentic Index	50.6	32.2
AA Intelligence	37.4	37.5
AA-LCR	68.3	76.7
AA-Omniscience	-0.1	5.4
AIME 2025	88	94.2
arc_agi_1	63.7	72.8
ARC-AGI-2	13.6	17.6
ARC-AGI-2 (Verified)	13.6	17.6
arena_elo	1456	1439
arena_text_factuality	1448	1445
Artificial Analysis Coding Index	52.1	49.4
browsecomp	24.1	50.8
coding_arena_elo	1392	1421
critpt	1.1	4.9
Fortress	17	25.7
frontiermath_tier_4	4.2	12.5
gdpval	40.4	25
GPQA Diamond	83.4	88.1
HLE	19.8	28.5
HLE (with tools)	32	42.7
hmmt_feb_2025	79.2	96.3
hmmt_nov_2025	81.7	91.7
IFBench	57.3	72.9
LiveCodeBench v6	64	87
matharena_visual_math_overall	75.8	76.9
MCP Atlas	59.5	50.1
MMLU-Pro	88.2	87
mmmlu	89.1	91
MMMU	79.6	85.4
MMMU-Pro	68.7	75.5
multichallenge	55.3	63.4
MultiNRC	35.8	49
OmniScience Accuracy	32.9	37.7
OmniScience Non-Hallucination	50.9	48.1
scicode	45	43.3
SEAL VISTA	48.8	43.8
simplebench	54.3	53.2
simpleqa_verified	23.6	48.9
swe_bench_bash	71.4	66
SWE-bench Verified	82	76.3
TauBench V3 - Banking	24.5	15.9
Terminal-Bench 2.0	50	47.6
Terminal-Bench 2.1	55.8	52.4
Terminal-Bench Hard	35.6	45.5
vectara_answer_rate	95.6	100
vectara_avg_summary_length	127.8	254.4
vectara_factual_consistency	88	89.1
vectara_hallucination_rate ↓	12	12.1
τ²-Bench	98	82.7
τ²-Bench Telecom (AA run)	78.1	81.9
τ³-Bench	19	14

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.