Which is better, Gemini 3 Pro or GPT-5.5?

Across 46 shared benchmarks tracked on Vector Wire, Gemini 3 Pro leads on 7 and GPT-5.5 leads on 38 — GPT-5.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Gemini 3 Pro vs GPT-5.5

GooglevsOpenAI46 shared benchmarks7–38 head-to-head

Benchmark	Gemini 3 Pro	GPT-5.5
AA Agentic Index	52	47.4
AA Intelligence	40.6	56.3
AA-LCR	73	79
AA-Omniscience	15.3	20.5
aime_2026	91.7	100
apexAgents	18.4	37.7
arc_agi_1	75	95
ARC-AGI-2	54	85
arena_elo	1485	1482
arena_text_factuality	1481	1482
arena_vision	1290	1283
Artificial Analysis Coding Index	46.5	74.9
browsecomp	59.2	84.4
coding_arena_elo	1438	1457
critpt	9.1	27.1
cybergym	39.9	81.8
Fortress	41.7	16.3
frontiermath_tier_4	18.8	35.4
gdpval	34.2	49.5
GDPval-AA (Elo)	1195	1769
GPQA Diamond	91.9	93.6
HLE	45.8	52.2
HLE (with tools)	45.8	52.2
hmmt_feb_2026	86.4	98.5
hmmt_nov_2025	93.3	96.5
IFBench	70.4	75.9
livebench	73.4	80.7
matharena_visual_math_overall	84.2	94.9
MCP Atlas	70.3	82.8
MMMU-Pro	81	83.2
OmniScience Accuracy	55.8	58
OmniScience Non-Hallucination	10	12
scicode	56.1	56.1
simplebench	76.4	69
simpleqa_verified	72.9	63.1
SWE-bench Pro	43.3	58.6
SWE-bench Verified	78	80.6
Terminal-Bench 2.0	54.2	82.7
Terminal-Bench Hard	56.9	60.6
tool_decathlon	36.4	55.6
toolathlon	36.4	55.6
vectara_answer_rate	99.4	100
vectara_avg_summary_length	101.9	129.6
vectara_factual_consistency	86.4	90.7
vectara_hallucination_rate ↓	13.6	9.3
τ²-Bench Telecom (AA run)	98	93.9

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.