Which is better, GPT-5.2 or Kimi K2.5?

Across 78 shared benchmarks tracked on Vector Wire, GPT-5.2 leads on 46 and Kimi K2.5 leads on 31 — GPT-5.2 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5.2 vs Kimi K2.5

OpenAIvsMoonshot78 shared benchmarks46–31 head-to-head

Benchmark	GPT-5.2	Kimi K2.5
AA Agentic Index	60.2	52.8
AA Intelligence	43.3	36
AA-LCR	79.3	73
AA-Omniscience	0.3	-7.3
AIME 2025	100	96.1
aime_2026	98.3	95.8
apexAgents	23	11.5
arc_agi_1	86.2	65.3
ARC-AGI-2	52.9	11.8
arena_elo	1437	1450
arena_text_factuality	1442	1445
arena_vision	1245	1247
Artificial Analysis Coding Index	48.7	46.8
browsecomp	65.8	74.9
BrowseComp (context management)	70	74.9
browsecomp_with_context_manager	65.8	74.9
browsecomp_zh	76.1	62.3
charxiv_rq	82.1	78.7
coding_arena_elo	1418	1436
critpt	11.6	3.1
deepsearchqa_f1	71.3	89
Fortress	17.5	41.1
frontiermath_tier_4	18.8	4.2
gdpval	48.3	38.3
GPQA Diamond	92.4	87.9
HLE	45.5	50.2
HLE (with tools)	45.5	51.8
HMMT 2025	99.4	95.4
hmmt_feb_2025	100	95.4
hmmt_feb_2026	97	87.1
hmmt_nov_2025	99.2	91.1
IFBench	75.4	70.2
imo_answer_bench	86.3	81.8
InfoVQA (val)	84	92.6
livebench	74.8	69.1
longbench_v2	54.5	61
LongVideoBench	76.5	79.8
matharena_visual_math_overall	86.5	80.6
mathvision	83	84.2
MathVista	82.8	90.1
MCP Atlas	68	64
MMLU-Pro	87	87.1
MMMU-Pro	79.5	78.5
MMVU	80.8	80.4
MotionBench	64.8	70.4
MultiNRC	42.2	35.2
OCRBench	80.7	92.3
OmniDocBench 1.5	85.7	88.8
OmniScience Accuracy	44.3	35.2
OmniScience Non-Hallucination	38.4	50
PaperBench	63.7	63.5
scicode	52.1	49
SEAL VISTA	46.6	41.9
Seal-0	45	57.4
simplebench	45.8	46.8
simpleqa_verified	38.9	36.9
SimpleVQA	55.8	71.2
swe_bench_bash	72.8	70.8
SWE-bench Multilingual	72	73
SWE-bench Pro	55.6	53.8
SWE-bench Verified	80	76.8
TauBench V3 - Banking	11.1	14.2
Terminal-Bench 2.0	54	50.8
Terminal-Bench Hard	62.2	34.8
tool_decathlon	46.3	27.8
toolathlon	46.3	27.8
vectara_answer_rate	100	92.2
vectara_avg_summary_length	186.3	112
vectara_factual_consistency	89.2	85.8
vectara_hallucination_rate ↓	10.8	14.2
Video-MME	86	87.4
VideoMMMU	85.9	86.6
WorldVQA	28	46.3
ZeroBench	9	9
ZeroBench (w/ tools)	7	11
τ²-Bench	85.5	85.4
τ²-Bench Telecom (AA run)	98.7	95.9
τ³-Bench	12	66

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.