Which is better, GPT-5.5 or Kimi K3?

Across 55 shared benchmarks tracked on Vector Wire, GPT-5.5 leads on 17 and Kimi K3 leads on 38 — 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-5.5 vs Kimi K3

OpenAIvsMoonshot55 shared benchmarks17–38 head-to-head

Benchmark	GPT-5.5	Kimi K3
AA Agentic Index	47.4	54.3
AA Intelligence	56.3	60
AA-LCR	79	82.7
AA-Omniscience	20.5	46
Agents' Last Exam	26.6	28.3
aime_2026	100	96.7
apex_agents	38.5	41
apexAgents	37.7	41.3
arc_agi_1	95	65.7
ARC-AGI-2	85	12.4
arena_elo	1482	1489
arena_text_factuality	1482	1471
Artificial Analysis Coding Index	74.9	76.2
browsecomp	84.4	91.2
coding_arena_elo	1457	1674
critpt	27.1	23.4
cybergym	81.8	80
DeepSWE	70	67.5
DeepSWE 1.1	67	69
Finance Agent v2	51.8	54.4
gdpval	49.5	59.4
GDPval-AA v2	1509	1687
GDPval-AA v2 (Elo)	1491	1686
GPQA Diamond	93.6	93.5
HLE	52.2	56
hmmt_feb_2026	98.5	97
itbenchSre	45.8	47.7
JobBench	38.3	54.3
livebench_agentic_coding	54	62.2
livebench_coding	82.2	81.5
livebench_data_analysis	81.6	78.7
livebench_instruction_following	70.7	71.4
livebench_language	87.4	85.5
livebench_math	95.9	84.4
livebench_reasoning	89.7	90.7
matharena_visual_math_overall	94.9	91.3
MCP Atlas	82.8	84.2
MCP Mark Verified	92.9	94.5
MLS Bench Lite	35.5	48.3
MMMU-Pro	83.2	83.4
officeqa_pro	60.9	63.3
OmniScience Accuracy	58	47.6
OmniScience Non-Hallucination	12	49.1
OSWorld-Verified	78.7	84.8
PostTrainBench	28.4	36.6
Program Bench	70.8	77.8
ResearchRubrics	64	76.2
scicode	56.1	58.7
simplebench	69	60.7
simpleqa_verified	63.1	42.7
SpreadsheetBench 2	29.1	34.8
SWE-Marathon	14	42
TauBench V3 - Banking	39	46
Terminal-Bench 2.1	84.3	88.3
toolathlon	55.6	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.