Which is better, Gemini 3.1 Pro or GPT-5.5?

Across 79 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 17 and GPT-5.5 leads on 60 — 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 3.1 Pro vs GPT-5.5

GooglevsOpenAI79 shared benchmarks17–60 head-to-head

Benchmark	Gemini 3.1 Pro	GPT-5.5
AA Agentic Index	23	47.4
AA Intelligence	47.7	56.3
AA-LCR	79	79
AA-Omniscience	32.9	20.5
aime_2026	98.3	100
AIRS-Bench	83	86
apex_agents	33.5	38.5
apexAgents	33.5	37.7
arc_agi_1	98	95
arc_agi_3	0.4	0.4
ARC-AGI-2	77.1	85
arena_elo	1486	1482
arena_text_factuality	1471	1482
arena_vision	1280	1283
Artificial Analysis Coding Index	68.8	74.9
browsecomp	85.9	84.4
coding_arena_elo	1447	1457
critpt	17.7	27.1
cybergym	38.8	81.8
DeepSWE	10	70
DeepSWE 1.1	12	67
Finance Agent v2	43	51.8
finance_agent	59.7	60
ForecastBench with search	61.9	64.3
Fortress	29.8	16.3
frontiermath_tier_4	16.7	35.4
gdpval	23.3	49.5
GDPval-AA (Elo)	1317	1769
GDPval-AA v2	962	1509
GPQA Diamond	94.3	93.6
HiL-Bench	35.3	39.7
HLE	51.4	52.2
HLE (with tools)	51.6	52.2
HLE Calibration	50.4	44.2
hmmt_feb_2026	94.7	98.5
hmmt_nov_2025	94.8	96.5
IFBench	77.1	75.9
itbenchSre	30.3	45.8
Legal Agent Benchmark	0	2.1
livebench	79.9	80.7
livebench_agentic_coding	44.1	54
livebench_coding	76.5	82.2
livebench_data_analysis	78.5	81.6
livebench_instruction_following	79.1	70.7
livebench_language	85.4	87.4
livebench_math	91	95.9
livebench_reasoning	84	89.7
matharena_visual_math_overall	89.4	94.9
MCP Atlas	78.2	82.8
MMMU-Pro	83	83.2
nl2repo	33.4	50.7
officeqa_pro	42.9	60.9
OmniScience Accuracy	55.3	58
OmniScience Non-Hallucination	50.1	12
OR-Bench (FRR)	2.5	6.7
OSWorld-Verified	76.2	78.7
PostTrainBench	21.6	28.4
Program Bench	39.5	70.8
Prophet Arena	0.1	0.1
scicode	59	56.1
simplebench	79.6	69
simpleqa_verified	77.3	63.1
SWE-bench Pro	54.2	58.6
SWE-bench Verified	80.6	80.6
SWE-Marathon	4	14
TauBench V3 - Banking	21.4	39
Terminal-Bench 2.0	68.5	82.7
Terminal-Bench 2.1	74	84.3
Terminal-Bench Hard	68.5	60.6
tool_decathlon	48.8	55.6
toolathlon	48.8	55.6
usamo_2026	74.4	98.2
vectara_answer_rate	99.4	100
vectara_avg_summary_length	107.7	129.6
vectara_factual_consistency	89.6	90.7
vectara_hallucination_rate ↓	10.4	9.3
wmdp	86.3	85.5
τ²-Bench Telecom (AA run)	99.3	93.9
τ³-Bench	67.1	31.3

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.