Which is better, Claude Sonnet 4.6 or GPT-5.6 Sol?

Across 45 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.6 leads on 4 and GPT-5.6 Sol leads on 41 — GPT-5.6 Sol 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 4.6 vs GPT-5.6 Sol

AnthropicvsOpenAI45 shared benchmarks4–41 head-to-head

Benchmark	Claude Sonnet 4.6	GPT-5.6 Sol
AA Agentic Index	61.6	57.8
AA Intelligence	48.4	61
AA-LCR	74	77.7
AA-Omniscience	12.2	22
arc_agi_1	86.5	97.5
ARC-AGI-2	60.4	92.5
arena_elo	1472	1482
arena_text_factuality	1460	1474
Artificial Analysis Coding Index	63	78.3
browsecomp	76.2	90.4
charxiv_rq	72.4	87.8
coding_arena_elo	1523	1619
critpt	3.1	32.3
DeepSWE 1.1	30	73
Finance Agent v2	51	53.8
GDP (Surge AI)	78.6	30.7
gdpval	54.8	61.1
GDPval-AA v2	1381	1748
GPQA Diamond	89.9	94.6
healthbench_professional	44.2	60.5
HLE	49	49.5
HLE (with tools)	46.8	58
IFBench	56.6	72.7
itbenchSre	39.8	56.2
livebench_agentic_coding	42.6	56.2
livebench_coding	79.3	83.9
livebench_data_analysis	78	79.8
livebench_instruction_following	63.2	71.8
livebench_language	76.1	87.7
livebench_math	87	96.2
livebench_reasoning	84.8	91.7
MCP Atlas	69.5	83.6
MMMU-Pro	75.6	84.6
officeqa_pro	53.4	63.2
OmniScience Accuracy	40.9	59.4
OmniScience Non-Hallucination	51.6	10.6
OSWorld-Verified	78.5	83
scicode	47	56.9
simpleqa_verified	29	71.6
SWE-bench Pro	58.1	64.6
TauBench V3 - Banking	34.4	44.3
Terminal-Bench 2.1	71.2	89.5
Terminal-Bench Hard	59.1	65.9
toolathlon	49.4	58
τ²-Bench Telecom (AA run)	97.9	85.1

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.