Which is better, Claude Opus 4.6 or Claude Opus 4.8?

Across 80 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 25 and Claude Opus 4.8 leads on 55 — 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 4.6 vs Claude Opus 4.8

AnthropicvsAnthropic80 shared benchmarks25–55 head-to-head

Benchmark	Claude Opus 4.6	Claude Opus 4.8
AA Agentic Index	67.6	49.4
AA Intelligence	44.9	57.3
AA-LCR	74.3	73
AA-Omniscience	13.7	28.8
AgentWorldBench - Android	61.7	61.5
AgentWorldBench - MCP	69.9	54.9
AgentWorldBench - OS	70.2	66.6
AgentWorldBench - Overall	57.8	56.6
AgentWorldBench - Search	29.3	35.1
AgentWorldBench - SWE	64.5	64.1
AgentWorldBench - Terminal	57.5	59.2
AgentWorldBench - Web	51.4	54.7
aime_2026	96.7	100
arc_agi_1	93	92.5
arc_agi_3	0.5	1.5
ARC-AGI-2	68.8	72.1
arena_elo	1497	1482
arena_text_factuality	1487	1464
arena_vision	1300	1280
Artificial Analysis Coding Index	48.1	74.3
baby_vision_with_python	38.4	81.2
browsecomp	86.8	88.5
BrowseComp (multi-agent)	86.6	88.5
BrowseComp (single-agent)	83.7	88.5
charxiv_reasoning_with_tools	84.7	89.9
charxiv_rq	69.1	80.5
coding_arena_elo	1545	1539
critpt	12.6	20.9
cybergym	73.8	83.1
deepsearchqa_f1	91.3	93.1
finance_agent	76.7	53.9
Fortress	20.5	18.2
frontiermath_tier_4	22.9	31.3
gdpval	55.9	54.2
GDPval-AA (Elo)	1619	1890
GPQA Diamond	91.3	93.6
HiL-Bench	38.3	35.3
HLE	53.1	57.9
HLE (with tools)	62.7	57.9
hmmt_feb_2026	96.2	96.7
hmmt_nov_2025	96.3	96.5
IFBench	62.5	62.2
imo_answer_bench	75.3	83.5
Legal Agent Benchmark	4.2	10
livebench	76.3	77.2
livebench_agentic_coding	49	50.5
livebench_coding	78.2	81.8
livebench_data_analysis	69.9	66
livebench_instruction_following	63.3	72
livebench_language	83.3	79.7
livebench_math	89.3	94.3
livebench_reasoning	88.7	89.2
matharena_visual_math_overall	72.3	81.6
mathvision	71.2	86.7
MCP Atlas	76.8	83.6
MMMU-Pro	77.3	78.9
nl2repo	49.8	69.7
officeqa	73.5	77.6
officeqa_pro	57.1	66.2
OmniScience Accuracy	47	48.8
OmniScience Non-Hallucination	37.2	60.7
OSWorld-Verified	72.7	83.4
scicode	52	53.5
screenspot_pro_no_tools	57.7	87.9
screenspot_pro_with_tools	83.1	87.9
simplebench	67.6	64.8
simpleqa_verified	46.5	39.5
swe_bench_multimodal	27.1	38.4
SWE-bench Multilingual	77.8	84.4
SWE-bench Pro	57.3	69.2
SWE-bench Verified	80.8	88.6
Terminal-Bench 2.0	65.4	74.6
Terminal-Bench Hard	65.4	58.3
tool_decathlon	47.2	59.9
toolathlon	56.8	59.9
Toolathlon Avg turns	16.9	24.5
Toolathlon Pass@∞	47.2	48.1
usamo_2026	66.2	96.7
τ²-Bench Telecom (AA run)	99.3	94.4
τ³-Bench	72.4	27.6

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.