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

Across 55 shared benchmarks tracked on Vector Wire, GPT-5 leads on 18 and Kimi K2.5 leads on 37 — Kimi K2.5 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 vs Kimi K2.5

OpenAIvsMoonshot55 shared benchmarks18–37 head-to-head

Benchmark	GPT-5	Kimi K2.5
AA Agentic Index	49.7	52.8
AA Intelligence	35.3	36
AA-LCR	76.3	73
AA-Omniscience	-8.7	-7.3
AIME 2025	95	96.1
arc_agi_1	56.2	65.3
ARC-AGI-2	7.5	11.8
arena_vision	1212	1247
Artificial Analysis Coding Index	38.9	46.8
browsecomp	54.9	74.9
browsecomp_zh	65	62.3
coding_arena_elo	1419	1436
critpt	5.7	3.1
Fortress	17	41.1
frontiermath_tier_4	12.5	4.2
gdpval	32.5	38.3
GPQA Diamond	87.3	87.9
HLE	28.5	50.2
HLE (with tools)	35.2	51.8
HMMT 2025	93.3	95.4
hmmt_feb_2025	88.3	95.4
hmmt_nov_2025	89.2	91.1
IFBench	73.1	70.2
imo_answer_bench	76	81.8
LiveCodeBench v6	87	85
matharena_visual_math_overall	78.8	80.6
MMLU-Pro	87.5	87.1
MMMU-Pro	78.4	78.5
multichallenge	71.1	61.4
MultiNRC	52.1	35.2
OJ-Bench (cpp)	56.2	57.4
OmniScience Accuracy	40.3	35.2
OmniScience Non-Hallucination	21	50
scicode	43	49
SEAL VISTA	49.7	41.9
Seal-0	51.4	57.4
simplebench	56.7	46.8
simpleqa_verified	50.6	36.9
swe_bench_bash	65	70.8
SWE-bench Multilingual	55.3	73
SWE-bench Pro	41.8	53.8
SWE-bench Verified	74.9	76.8
TauBench V3 - Banking	22.1	14.2
Terminal-Bench 2.0	35.2	50.8
Terminal-Bench 2.1	35.2	45.7
Terminal-Bench Hard	37.9	34.8
vectara_answer_rate	99.9	92.2
vectara_avg_summary_length	162.7	112
vectara_factual_consistency	85.3	85.8
vectara_hallucination_rate ↓	14.7	14.2
VideoMMMU	84.6	86.6
xbench_deepsearch	77.8	76.7
τ²-Bench	85	85.4
τ²-Bench Telecom (AA run)	86.5	95.9
τ³-Bench	19.6	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.