Which is better, GPT-5.4 or Kimi K3?

Across 45 shared benchmarks tracked on Vector Wire, GPT-5.4 leads on 11 and Kimi K3 leads on 33 — Kimi K3 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5.4 vs Kimi K3

OpenAIvsMoonshot45 shared benchmarks11–33 head-to-head

Benchmark	GPT-5.4	Kimi K3
AA Agentic Index	58.2	54.3
AA Intelligence	53.1	60
AA-LCR	77.7	82.7
AA-Omniscience	5.8	46
aime_2026	99.2	96.7
apexAgents	33.3	41.3
arc_agi_1	93.7	65.7
ARC-AGI-2	74	12.4
arena_elo	1477	1489
arena_text_factuality	1476	1471
Artificial Analysis Coding Index	71.1	76.2
baby_vision	49.7	85.7
baby_vision_with_python	80.2	85.7
browsecomp	82.7	91.2
charxiv_rq	82.8	84.8
coding_arena_elo	1457	1674
critpt	23.4	23.4
cybergym	66.3	80
deepsearchqa_f1	78.6	95
DeepSWE 1.1	52	69
gdpval	50.1	59.4
GPQA Diamond	93	93.5
HLE	43.7	56
hmmt_feb_2026	97.7	97
itbenchSre	34.5	47.7
livebench_agentic_coding	53.8	62.2
livebench_coding	77.5	81.5
livebench_data_analysis	79.3	78.7
livebench_instruction_following	70.2	71.4
livebench_language	82.6	85.5
livebench_math	94.2	84.4
livebench_reasoning	88.1	90.7
matharena_visual_math_overall	92.5	91.3
mathvision	92	97.8
MCP Atlas	70.6	84.2
MMMU-Pro	81.2	83.4
officeqa_pro	51.1	63.3
OmniScience Accuracy	50.9	47.6
OmniScience Non-Hallucination	17.4	49.1
OSWorld-Verified	75	84.8
scicode	56.6	58.7
simpleqa_verified	45.3	42.7
TauBench V3 - Banking	39.6	46
Terminal-Bench 2.1	78.3	88.3
toolathlon	54.6	73.2

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.