Which is better, Gemini 3 Flash Preview or GPT-5.4?

Across 47 shared benchmarks tracked on Vector Wire, Gemini 3 Flash Preview leads on 6 and GPT-5.4 leads on 40 — GPT-5.4 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 Flash Preview vs GPT-5.4

GooglevsOpenAI47 shared benchmarks6–40 head-to-head

Benchmark	Gemini 3 Flash Preview	GPT-5.4
AA Agentic Index	49.7	58.2
AA Intelligence	38.7	53.1
AA-LCR	73	77.7
AA-Omniscience	10.1	5.8
aime_2026	96.7	99.2
apexAgents	27.7	33.3
arc_agi_1	84.7	93.7
ARC-AGI-2	33.6	74
arena_elo	1473	1477
arena_text_factuality	1452	1476
arena_vision	1260	1277
Artificial Analysis Coding Index	42.6	71.1
charxiv_rq	80.3	82.8
coding_arena_elo	1387	1457
critpt	8.6	23.4
frontiermath_tier_4	4.2	27.1
gdpval	35.2	50.1
GDPval-AA (Elo)	1204	1674
GPQA Diamond	90.4	93
HLE	43.5	43.7
hmmt_feb_2026	89.4	97.7
hmmt_nov_2025	93.3	95.8
IFBench	78	73.9
Legal Agent Benchmark	0	0.4
livebench	72.4	80.3
matharena_visual_math_overall	85.8	92.5
MCP Atlas	62	70.6
MMMU-Pro	81.2	81.2
OmniDocBench 1.5	12.1	89.1
OmniScience Accuracy	53.4	50.9
OmniScience Non-Hallucination	7.6	17.4
OSWorld-Verified	65.1	75
scicode	50.6	56.6
simpleqa_verified	67.4	45.3
SWE-bench Multilingual	72.7	71.7
SWE-bench Pro	49.6	59.1
TauBench V3 - Banking	20.8	39.6
Terminal-Bench 2.0	47.6	75.1
Terminal-Bench 2.1	58	78.3
Terminal-Bench Hard	38.6	57.6
toolathlon	49.4	54.6
vectara_answer_rate	99.8	99.9
vectara_avg_summary_length	90.2	81.7
vectara_factual_consistency	86.5	93
vectara_hallucination_rate ↓	13.5	7
τ²-Bench Telecom (AA run)	80.4	87.1
τ³-Bench	17.5	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.