Which is better, Claude Opus 4.8 or Gemini 3 Flash Preview?

Across 45 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 38 and Gemini 3 Flash Preview leads on 6 — 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 Flash Preview

AnthropicvsGoogle45 shared benchmarks38–6 head-to-head

Benchmark	Claude Opus 4.8	Gemini 3 Flash Preview
AA Agentic Index	49.4	49.7
AA Intelligence	57.3	38.7
AA-LCR	73	73
AA-Omniscience	28.8	10.1
aime_2026	100	96.7
arc_agi_1	92.5	84.7
ARC-AGI-2	72.1	33.6
arena_elo	1482	1473
arena_text_factuality	1464	1452
arena_vision	1280	1260
Artificial Analysis Coding Index	74.3	42.6
charxiv_rq	80.5	80.3
coding_arena_elo	1539	1387
critpt	20.9	8.6
Finance Agent v2	53.9	42.6
frontiermath_tier_4	31.3	4.2
gdpval	54.2	35.2
GDPval-AA (Elo)	1890	1204
GPQA Diamond	93.6	90.4
HLE	57.9	43.5
hmmt_feb_2026	96.7	89.4
hmmt_nov_2025	96.5	93.3
IFBench	62.2	78
Legal Agent Benchmark	10	0
livebench	77.2	72.4
matharena_visual_math_overall	81.6	85.8
MCP Atlas	83.6	62
MMMU-Pro	78.9	81.2
OmniScience Accuracy	48.8	53.4
OmniScience Non-Hallucination	60.7	7.6
OSWorld-Verified	83.4	65.1
scicode	53.5	50.6
screenspot_pro_no_tools	87.9	69.1
simplebench	64.8	61.1
simpleqa_verified	39.5	67.4
SWE-bench Multilingual	84.4	72.7
SWE-bench Pro	69.2	49.6
SWE-bench Verified	88.6	78
TauBench V3 - Banking	34.2	20.8
Terminal-Bench 2.0	74.6	47.6
Terminal-Bench 2.1	85	58
Terminal-Bench Hard	58.3	38.6
toolathlon	59.9	49.4
τ²-Bench Telecom (AA run)	94.4	80.4
τ³-Bench	27.6	17.5

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.