Which is better, GPT-5.5 or GPT-5.6 Sol?

Across 64 shared benchmarks tracked on Vector Wire, GPT-5.5 leads on 11 and GPT-5.6 Sol leads on 51 — 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-5.5 vs GPT-5.6 Sol

OpenAIvsOpenAI64 shared benchmarks11–51 head-to-head

Benchmark	GPT-5.5	GPT-5.6 Sol
AA Agentic Index	47.4	57.8
AA Intelligence	56.3	61
AA-LCR	79	77.7
AA-Omniscience	20.5	22
Agents' Last Exam	26.6	52.7
aime_2026	100	99.9
apex_agents	38.5	56.7
arc_agi_1	95	97.5
arc_agi_3	0.4	7.8
ARC-AGI-2	85	92.5
arena_elo	1482	1482
arena_text_factuality	1482	1474
Artificial Analysis Coding Index	74.9	78.3
browsecomp	84.4	90.4
coding_arena_elo	1457	1619
critpt	27.1	32.3
cybergym	81.8	83.6
DeepSWE	70	73
DeepSWE 1.1	67	73
DNA sequence design for transcription factor binding	13.8	13.7
Finance Agent v2	51.8	53.8
GDP (Surge AI)	16.7	30.7
gdpval	49.5	61.1
GDPval-AA v2	1509	1748
GDPval-AA v2 (Elo)	1491	1736
GPQA Diamond	93.6	94.6
Hard-negative protein binding prediction	1.5	7.6
HealthBench	56.5	57
healthbench_professional	51.8	60.5
HiL-Bench	39.7	32.3
HLE	52.2	49.5
HLE (with tools)	52.2	58
IFBench	75.9	72.7
itbenchSre	45.8	56.2
JobBench	38.3	45.4
livebench_agentic_coding	54	56.2
livebench_coding	82.2	83.9
livebench_data_analysis	81.6	79.8
livebench_instruction_following	70.7	71.8
livebench_language	87.4	87.7
livebench_math	95.9	96.2
livebench_reasoning	89.7	91.7
MCP Atlas	82.8	83.6
MCP Mark Verified	92.9	92.9
MLS Bench Lite	35.5	46.2
MMMU-Pro	83.2	84.6
officeqa_pro	60.9	63.2
OmniScience Accuracy	58	59.4
OmniScience Non-Hallucination	12	10.6
OSWorld-Verified	78.7	83
PostTrainBench	28.4	34.6
Program Bench	70.8	77.6
ResearchRubrics	64	73.8
scicode	56.1	56.9
simplebench	69	64.8
simpleqa_verified	63.1	71.6
SpreadsheetBench 2	29.1	32.4
SWE-bench Pro	58.6	64.6
SWE-Marathon	14	39
TauBench V3 - Banking	39	44.3
Terminal-Bench 2.1	84.3	89.5
Terminal-Bench Hard	60.6	65.9
toolathlon	55.6	58
τ²-Bench Telecom (AA run)	93.9	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.