Which is better, Claude Opus 4.8 or Gemini 3.1 Pro?

Across 86 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 55 and Gemini 3.1 Pro leads on 29 — 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 Gemini 3.1 Pro

AnthropicvsGoogle86 shared benchmarks55–29 head-to-head

Benchmark	Claude Opus 4.8	Gemini 3.1 Pro
AA Agentic Index	49.4	23
AA Intelligence	57.3	47.7
AA-LCR	73	79
AA-Omniscience	28.8	32.9
AgentWorldBench - Android	61.5	61.4
AgentWorldBench - MCP	54.9	59.1
AgentWorldBench - OS	66.6	66.9
AgentWorldBench - Overall	56.6	54.6
AgentWorldBench - Search	35.1	30.2
AgentWorldBench - SWE	64.1	59.1
AgentWorldBench - Terminal	59.2	52.5
AgentWorldBench - Web	54.7	52.8
aime_2026	100	98.3
AIRS-Bench	84	83
apex_agents	39.4	33.5
arc_agi_1	92.5	98
arc_agi_3	1.5	0.4
ARC-AGI-2	72.1	77.1
arena_elo	1482	1486
arena_text_factuality	1464	1471
arena_vision	1280	1280
Artificial Analysis Coding Index	74.3	68.8
baby_vision_with_python	81.2	68.3
browsecomp	88.5	85.9
charxiv_reasoning_with_tools	89.9	83.2
charxiv_rq	80.5	89.9
coding_arena_elo	1539	1447
critpt	20.9	17.7
cybergym	83.1	38.8
deepsearchqa_f1	93.1	81.9
DeepSWE	59	10
DeepSWE 1.1	59	12
Finance Agent v2	53.9	43
finance_agent	53.9	59.7
Fortress	18.2	29.8
frontiermath_tier_4	31.3	16.7
gdpval	54.2	23.3
GDPval-AA (Elo)	1890	1317
GDPval-AA v2	1593	962
GPQA Diamond	93.6	94.3
HiL-Bench	35.3	35.3
HLE	57.9	51.4
HLE (with tools)	57.9	51.6
HLE Calibration	26.5	50.4
hmmt_feb_2026	96.7	94.7
hmmt_nov_2025	96.5	94.8
IFBench	62.2	77.1
imo_answer_bench	83.5	91
Legal Agent Benchmark	10	0
livebench	77.2	79.9
livebench_agentic_coding	50.5	44.1
livebench_coding	81.8	76.5
livebench_data_analysis	66	78.5
livebench_instruction_following	72	79.1
livebench_language	79.7	85.4
livebench_math	94.3	91
livebench_reasoning	89.2	84
matharena_visual_math_overall	81.6	89.4
mathvision	86.7	89.8
MCP Atlas	83.6	78.2
MMMU-Pro	78.9	83
nl2repo	69.7	33.4
officeqa_pro	66.2	42.9
OmniScience Accuracy	48.8	55.3
OmniScience Non-Hallucination	60.7	50.1
OR-Bench (FRR)	3.3	2.5
OSWorld-Verified	83.4	76.2
PostTrainBench	37.2	21.6
Program Bench	71.9	39.5
scicode	53.5	59
simplebench	64.8	79.6
simpleqa_verified	39.5	77.3
SWE-bench Multilingual	84.4	76.9
SWE-bench Pro	69.2	54.2
SWE-bench Verified	88.6	80.6
SWE-Marathon	40	4
Tau 3 Banking	27.6	16.5
TauBench V3 - Banking	34.2	21.4
Terminal-Bench 2.0	74.6	68.5
Terminal-Bench 2.1	85	74
Terminal-Bench Hard	58.3	68.5
tool_decathlon	59.9	48.8
toolathlon	59.9	48.8
usamo_2026	96.7	74.4
τ²-Bench Telecom (AA run)	94.4	99.3
τ³-Bench	27.6	67.1

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.