Which is better, GPT-5.2 or GPT-5.6 Sol?

Across 43 shared benchmarks tracked on Vector Wire, GPT-5.2 leads on 5 and GPT-5.6 Sol leads on 38 — 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 —

GPT-5.2 vs GPT-5.6 Sol

OpenAIvsOpenAI43 shared benchmarks5–38 head-to-head

Benchmark	GPT-5.2	GPT-5.6 Sol
AA Agentic Index	60.2	57.8
AA Intelligence	43.3	61
AA-LCR	79.3	77.7
AA-Omniscience	0.3	22
aime_2026	98.3	99.9
arc_agi_1	86.2	97.5
ARC-AGI-2	52.9	92.5
arena_elo	1437	1482
arena_text_factuality	1442	1474
Artificial Analysis Coding Index	48.7	78.3
browsecomp	65.8	90.4
charxiv_rq	82.1	87.8
coding_arena_elo	1418	1619
critpt	11.6	32.3
gdpval	48.3	61.1
GPQA Diamond	92.4	94.6
HLE	45.5	49.5
HLE (with tools)	45.5	58
IFBench	75.4	72.7
livebench_agentic_coding	50.3	56.2
livebench_coding	76.1	83.9
livebench_data_analysis	78.2	79.8
livebench_instruction_following	61.8	71.8
livebench_language	79.8	87.7
livebench_math	93.2	96.2
livebench_reasoning	83.2	91.7
longbench_v2	54.5	67.1
mathvision	83	95.8
MCP Atlas	68	83.6
MMMU-Pro	79.5	84.6
MMVU	80.8	81.2
OmniScience Accuracy	44.3	59.4
OmniScience Non-Hallucination	38.4	10.6
PaperBench	63.7	90.5
scicode	52.1	56.9
simplebench	45.8	64.8
simpleqa_verified	38.9	71.6
SWE-bench Pro	55.6	64.6
TauBench V3 - Banking	11.1	44.3
Terminal-Bench Hard	62.2	65.9
toolathlon	46.3	58
Video-MME	86	89.5
τ²-Bench Telecom (AA run)	98.7	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.