Which is better, Gemini 3.5 Flash or GPT-5.4?

Across 50 shared benchmarks tracked on Vector Wire, Gemini 3.5 Flash leads on 25 and GPT-5.4 leads on 24 — Gemini 3.5 Flash 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-5.4

GooglevsOpenAI50 shared benchmarks25–24 head-to-head

Benchmark	Gemini 3.5 Flash	GPT-5.4
AA Agentic Index	70.4	58.2
AA Intelligence	52	53.1
AA-LCR	81	77.7
AA-Omniscience	21.2	5.8
aime_2026	95	99.2
apexAgents	47.1	33.3
arc_agi_1	92.5	93.7
ARC-AGI-2	72.1	74
arena_elo	1477	1477
arena_text_factuality	1466	1476
Artificial Analysis Coding Index	70.1	71.1
charxiv_rq	84.2	82.8
coding_arena_elo	1490	1457
critpt	13.1	23.4
DeepSWE 1.1	37	52
finance_agent	57.9	57.2
frontiermath_tier_4	14.6	27.1
gdpval	57.8	50.1
GDPval-AA (Elo)	1656	1674
GPQA Diamond	92.2	93
HiL-Bench	27.7	9.7
HLE	42.7	43.7
hmmt_feb_2026	95.5	97.7
IFBench	76.3	73.9
itbenchSre	40.3	34.5
Legal Agent Benchmark	0.8	0.4
livebench	75	80.3
livebench_agentic_coding	49	53.8
livebench_coding	78.2	77.5
livebench_data_analysis	64.9	79.3
livebench_instruction_following	75.6	70.2
livebench_language	84.6	82.6
livebench_math	88.2	94.2
livebench_reasoning	82	88.1
matharena_visual_math_overall	89.9	92.5
MCP Atlas	83.6	70.6
MMMU-Pro	84.3	81.2
OmniScience Accuracy	51.4	50.9
OmniScience Non-Hallucination	38.2	17.4
OSWorld-Verified	78.4	75
scicode	53.1	56.6
simpleqa_verified	68.4	45.3
SWE-bench Pro	55.1	59.1
TauBench V3 - Banking	32.2	39.6
Terminal-Bench 2.0	76.2	75.1
Terminal-Bench 2.1	78.7	78.3
Terminal-Bench Hard	46.2	57.6
toolathlon	56.5	54.6
τ²-Bench Telecom (AA run)	95.6	87.1
τ³-Bench	25.4	72.9

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.