Which is better, GPT-5.1 or Kimi K2.5?

Across 49 shared benchmarks tracked on Vector Wire, GPT-5.1 leads on 28 and Kimi K2.5 leads on 21 — GPT-5.1 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5.1 vs Kimi K2.5

OpenAIvsMoonshot49 shared benchmarks28–21 head-to-head

Benchmark	GPT-5.1	Kimi K2.5
AA Agentic Index	32.2	52.8
AA Intelligence	37.5	36
AA-LCR	76.7	73
AA-Omniscience	5.4	-7.3
AIME 2025	94.2	96.1
arc_agi_1	72.8	65.3
ARC-AGI-2	17.6	11.8
arena_elo	1439	1450
arena_text_factuality	1445	1445
arena_vision	1250	1247
Artificial Analysis Coding Index	49.4	46.8
browsecomp	50.8	74.9
coding_arena_elo	1421	1436
critpt	4.9	3.1
Fortress	25.7	41.1
frontiermath_tier_4	12.5	4.2
gdpval	25	38.3
GPQA Diamond	88.1	87.9
HLE	28.5	50.2
HLE (with tools)	42.7	51.8
hmmt_feb_2025	96.3	95.4
hmmt_nov_2025	91.7	91.1
IFBench	72.9	70.2
LiveCodeBench v6	87	85
matharena_visual_math_overall	76.9	80.6
MCP Atlas	50.1	64
MMLU-Pro	87	87.1
MMMU-Pro	75.5	78.5
multichallenge	63.4	61.4
MultiNRC	49	35.2
OmniScience Accuracy	37.7	35.2
OmniScience Non-Hallucination	48.1	50
scicode	43.3	49
SEAL VISTA	43.8	41.9
simplebench	53.2	46.8
simpleqa_verified	48.9	36.9
swe_bench_bash	66	70.8
SWE-bench Verified	76.3	76.8
TauBench V3 - Banking	15.9	14.2
Terminal-Bench 2.0	47.6	50.8
Terminal-Bench 2.1	52.4	45.7
Terminal-Bench Hard	45.5	34.8
vectara_answer_rate	100	92.2
vectara_avg_summary_length	254.4	112
vectara_factual_consistency	89.1	85.8
vectara_hallucination_rate ↓	12.1	14.2
τ²-Bench	82.7	85.4
τ²-Bench Telecom (AA run)	81.9	95.9
τ³-Bench	14	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.