Which is better, GPT-5.5 or Kimi K2.5?

Across 52 shared benchmarks tracked on Vector Wire, GPT-5.5 leads on 47 and Kimi K2.5 leads on 5 — GPT-5.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5.5 vs Kimi K2.5

OpenAIvsMoonshot52 shared benchmarks47–5 head-to-head

Benchmark	GPT-5.5	Kimi K2.5
AA Agentic Index	47.4	52.8
AA Intelligence	56.3	36
AA-LCR	79	73
AA-Omniscience	20.5	-7.3
aime_2026	100	95.8
apexAgents	37.7	11.5
arc_agi_1	95	65.3
ARC-AGI-2	85	11.8
arena_elo	1482	1450
arena_text_factuality	1482	1445
arena_vision	1283	1247
Artificial Analysis Coding Index	74.9	46.8
browsecomp	84.4	74.9
coding_arena_elo	1457	1436
critpt	27.1	3.1
cybergym	81.8	41.3
Fortress	16.3	41.1
frontiermath_tier_4	35.4	4.2
gdpval	49.5	38.3
GDPval-AA v2	1509	1009
GPQA Diamond	93.6	87.9
HLE	52.2	50.2
HLE (with tools)	52.2	51.8
hmmt_feb_2026	98.5	87.1
hmmt_nov_2025	96.5	91.1
IFBench	75.9	70.2
livebench	80.7	69.1
matharena_visual_math_overall	94.9	80.6
MCP Atlas	82.8	64
MMMU-Pro	83.2	78.5
nl2repo	50.7	32
OmniScience Accuracy	58	35.2
OmniScience Non-Hallucination	12	50
OSWorld-Verified	78.7	63.3
ResearchRubrics	64	59.5
scicode	56.1	49
simplebench	69	46.8
simpleqa_verified	63.1	36.9
SWE-bench Pro	58.6	53.8
SWE-bench Verified	80.6	76.8
TauBench V3 - Banking	39	14.2
Terminal-Bench 2.0	82.7	50.8
Terminal-Bench 2.1	84.3	45.7
Terminal-Bench Hard	60.6	34.8
tool_decathlon	55.6	27.8
toolathlon	55.6	27.8
vectara_answer_rate	100	92.2
vectara_avg_summary_length	129.6	112
vectara_factual_consistency	90.7	85.8
vectara_hallucination_rate ↓	9.3	14.2
τ²-Bench Telecom (AA run)	93.9	95.9
τ³-Bench	31.3	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.