Which is better, Claude Opus 4.7 or Gemini 3 Flash Preview?

Across 53 shared benchmarks tracked on Vector Wire, Claude Opus 4.7 leads on 43 and Gemini 3 Flash Preview leads on 10 — Claude Opus 4.7 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 4.7 vs Gemini 3 Flash Preview

AnthropicvsGoogle53 shared benchmarks43–10 head-to-head

Benchmark	Claude Opus 4.7	Gemini 3 Flash Preview
AA Agentic Index	64.6	49.7
AA Intelligence	55	38.7
AA-LCR	75.3	73
AA-Omniscience	27.3	10.1
aime_2026	95.8	96.7
arc_agi_1	92	84.7
ARC-AGI-2	75.8	33.6
arena_elo	1494	1473
arena_text_factuality	1474	1452
arena_vision	1304	1260
Artificial Analysis Coding Index	73.6	42.6
Blueprint-Bench 2	24.5	0
charxiv_rq	82.1	80.3
coding_arena_elo	1557	1387
critpt	12	8.6
Finance Agent v2	51.5	42.6
frontiermath_tier_4	22.9	4.2
gaia_overall	28	68.4
gdpval	58.6	35.2
GDPval-AA (Elo)	1753	1204
GPQA Diamond	94.2	90.4
harmbench	17.5	70
HLE	54.7	43.5
hmmt_feb_2026	93.9	89.4
Humanity’s Last Exam	46.9	33.7
IFBench	58.6	78
Legal Agent Benchmark	7.1	0
livebench	76.9	72.4
MCP Atlas	79.1	62
mmmlu	91.5	91.8
MMMU-Pro	78.8	81.2
mrcr_v2_8needle_128k_average	59.3	67.2
OmniScience Accuracy	48.9	53.4
OmniScience Non-Hallucination	57.7	7.6
OSWorld-Verified	82.8	65.1
scicode	54.5	50.6
screenspot_pro_no_tools	79.5	69.1
simplebench	61.7	61.1
simpleqa_verified	50.6	67.4
SWE-bench Multilingual	80.5	72.7
SWE-bench Pro	64.3	49.6
SWE-bench Verified	87.6	78
TauBench V3 - Banking	34.6	20.8
Terminal-Bench 2.0	69.4	47.6
Terminal-Bench 2.1	83.1	58
Terminal-Bench Hard	54.5	38.6
toolathlon	59.3	49.4
vectara_answer_rate	98	99.8
vectara_avg_summary_length	149.1	90.2
vectara_factual_consistency	88	86.5
vectara_hallucination_rate ↓	12	13.5
τ²-Bench Telecom (AA run)	88.6	80.4
τ³-Bench	28.9	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.