Which is better, Claude Opus 4.8 or Kimi K2.6?

Across 65 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 55 and Kimi K2.6 leads on 10 — Claude Opus 4.8 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 K2.6

AnthropicvsMoonshot65 shared benchmarks55–10 head-to-head

Benchmark	Claude Opus 4.8	Kimi K2.6
AA Agentic Index	49.4	58.7
AA Intelligence	57.3	45.1
AA-LCR	73	76.7
AA-Omniscience	28.8	6.4
AgentWorldBench - Android	61.5	58.9
AgentWorldBench - MCP	54.9	65.2
AgentWorldBench - OS	66.6	60.8
AgentWorldBench - Overall	56.6	53.4
AgentWorldBench - Search	35.1	27.5
AgentWorldBench - SWE	64.1	58.8
AgentWorldBench - Terminal	59.2	52.5
AgentWorldBench - Web	54.7	50.2
aime_2026	100	96.4
apex_agents	39.4	27.9
arena_elo	1482	1461
arena_text_factuality	1464	1455
arena_vision	1280	1264
Artificial Analysis Coding Index	74.3	61.8
baby_vision_with_python	81.2	68.5
browsecomp	88.5	86.3
charxiv_rq	80.5	86.7
coding_arena_elo	1539	1509
critpt	20.9	8
deepsearchqa_f1	93.1	92.5
Finance Agent v2	53.9	44.9
frontiermath_tier_4	31.3	14.6
gdpval	54.2	41.4
GDPval-AA (Elo)	1890	1482
GDPval-AA v2	1593	1190
GPQA Diamond	93.6	91.1
HiL-Bench	35.3	18.7
HLE	57.9	37.5
HLE (with tools)	57.9	54
hmmt_feb_2026	96.7	94.7
IFBench	62.2	76
imo_answer_bench	83.5	86
livebench	77.2	72.2
livebench_agentic_coding	50.5	46.9
livebench_coding	81.8	78.6
livebench_data_analysis	66	65.1
livebench_instruction_following	72	64.4
livebench_language	79.7	75.1
livebench_math	94.3	84.3
livebench_reasoning	89.2	79.4
mathvision	86.7	93.2
MCP Atlas	83.6	68.1
MMMU-Pro	78.9	80.1
OmniScience Accuracy	48.8	32.8
OmniScience Non-Hallucination	60.7	60.7
OSWorld-Verified	83.4	73.1
scicode	53.5	53.5
simpleqa_verified	39.5	38.7
SWE-bench Multilingual	84.4	76.7
SWE-bench Pro	69.2	58.6
SWE-bench Verified	88.6	80.2
Tau 3 Banking	27.6	20.6
TauBench V3 - Banking	34.2	23.3
Terminal-Bench 2.0	74.6	66.7
Terminal-Bench 2.1	85	65.9
Terminal-Bench Hard	58.3	43.9
toolathlon	59.9	50
usamo_2026	96.7	51.2
WideSearch	72.9	80.8
τ²-Bench Telecom (AA run)	94.4	95.9
τ³-Bench	27.6	20.6

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.