Which is better, Gemini 3.5 Flash or GPT-5.5?

Across 54 shared benchmarks tracked on Vector Wire, Gemini 3.5 Flash leads on 16 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 3.5 Flash vs GPT-5.5

GooglevsOpenAI54 shared benchmarks16–38 head-to-head

Benchmark	Gemini 3.5 Flash	GPT-5.5
AA Agentic Index	70.4	47.4
AA Intelligence	52	56.3
AA-LCR	81	79
AA-Omniscience	21.2	20.5
aime_2026	95	100
apexAgents	47.1	37.7
arc_agi_1	92.5	95
ARC-AGI-2	72.1	85
arena_elo	1477	1482
arena_text_factuality	1466	1482
Artificial Analysis Coding Index	70.1	74.9
coding_arena_elo	1490	1457
critpt	13.1	27.1
DeepSWE	37	70
DeepSWE 1.1	37	67
Finance Agent v2	57.9	51.8
finance_agent	57.9	60
frontiermath_tier_4	14.6	35.4
gdpval	57.8	49.5
GDPval-AA (Elo)	1656	1769
GDPval-AA v2	1357	1509
GPQA Diamond	92.2	93.6
HiL-Bench	27.7	39.7
HLE	42.7	52.2
hmmt_feb_2026	95.5	98.5
IFBench	76.3	75.9
itbenchSre	40.3	45.8
Legal Agent Benchmark	0.8	2.1
Legal Agent Benchmark (Harvey's Held-Out Set)	0.8	2.1
livebench	75	80.7
livebench_agentic_coding	49	54
livebench_coding	78.2	82.2
livebench_data_analysis	64.9	81.6
livebench_instruction_following	75.6	70.7
livebench_language	84.6	87.4
livebench_math	88.2	95.9
livebench_reasoning	82	89.7
matharena_visual_math_overall	89.9	94.9
MCP Atlas	83.6	82.8
MMMU-Pro	84.3	83.2
OmniScience Accuracy	51.4	58
OmniScience Non-Hallucination	38.2	12
OSWorld-Verified	78.4	78.7
scicode	53.1	56.1
simplebench	76.7	69
simpleqa_verified	68.4	63.1
SWE-bench Pro	55.1	58.6
TauBench V3 - Banking	32.2	39
Terminal-Bench 2.0	76.2	82.7
Terminal-Bench 2.1	78.7	84.3
Terminal-Bench Hard	46.2	60.6
toolathlon	56.5	55.6
τ²-Bench Telecom (AA run)	95.6	93.9
τ³-Bench	25.4	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.