Which is better, Claude Opus 4.8 or Kimi K3?

Across 71 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 16 and Kimi K3 leads on 55 — 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 —

Claude Opus 4.8 vs Kimi K3

AnthropicvsMoonshot71 shared benchmarks16–55 head-to-head

Benchmark	Claude Opus 4.8	Kimi K3
AA Agentic Index	49.4	54.3
AA Intelligence	57.3	60
AA-LCR	73	82.7
AA-Omniscience	28.8	46
Agents' Last Exam	27	28.3
aime_2026	100	96.7
apex_agents	39.4	41
arc_agi_1	92.5	65.7
ARC-AGI-2	72.1	12.4
arena_elo	1482	1489
arena_text_factuality	1464	1471
Artificial Analysis Coding Index	74.3	76.2
AutomationBench (Public)	27.2	30.8
baby_vision_with_python	81.2	85.7
browsecomp	88.5	91.2
charxiv_rq	80.5	84.8
coding_arena_elo	1539	1674
CorpFin v2	66.7	71.6
critpt	20.9	23.4
cybergym	83.1	80
deepsearchqa_f1	93.1	95
DeepSWE	59	67.5
DeepSWE 1.1	59	69
DSBench-FullStack	71.6	73.7
DSBench-Hard	71.7	63
Finance Agent v2	53.9	54.4
gdpval	54.2	59.4
GDPval-AA v2	1593	1687
GDPval-AA v2 (Elo)	1593	1686
GPQA Diamond	93.6	93.5
HLE	57.9	56
HLE (wo / w tools)	57.9	56
hmmt_feb_2026	96.7	97
JobBench	48.4	54.3
Legal Research Bench	43.8	44.2
livebench_agentic_coding	50.5	62.2
livebench_coding	81.8	81.5
livebench_data_analysis	66	78.7
livebench_instruction_following	72	71.4
livebench_language	79.7	85.5
livebench_math	94.3	84.4
livebench_reasoning	89.2	90.7
matharena_visual_math_overall	81.6	91.3
mathvision	86.7	97.8
MCP Atlas	83.6	84.2
MCP Mark Verified	76.4	94.5
MLS Bench Lite	42.8	48.3
MMMU-Pro	78.9	83.4
MMVU	79.2	82.1
officeqa_pro	66.2	63.3
OmniDocBench	87.9	91.1
OmniScience Accuracy	48.8	47.6
OmniScience Non-Hallucination	60.7	49.1
OSWorld 2.0	55.7	58.3
OSWorld-Verified	83.4	84.8
PerceptionBench	47.2	58.5
PostTrainBench	37.2	36.6
Program Bench	71.9	77.8
ResearchRubrics	73.5	76.2
SaaS-Bench	56.1	60.1
scicode	53.5	58.7
simplebench	64.8	60.7
simpleqa_verified	39.5	42.7
SpreadsheetBench 2	31.6	34.8
SWE-Marathon	40	42
Tau 3 Banking	27.6	33.4
TauBench V3 - Banking	34.2	46
Terminal-Bench 2.1	85	88.3
toolathlon	59.9	73.2
Video-MME	86	90
WorldVQA ForceAnswer	39.1	51

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.