Which is better, Claude Opus 4.6 or Gemini 3.5 Flash?

Across 49 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 21 and Gemini 3.5 Flash leads on 26 — Gemini 3.5 Flash 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 Gemini 3.5 Flash

AnthropicvsGoogle49 shared benchmarks21–26 head-to-head

Benchmark	Claude Opus 4.6	Gemini 3.5 Flash
AA Agentic Index	67.6	70.4
AA Intelligence	44.9	52
AA-LCR	74.3	81
AA-Omniscience	13.7	21.2
aime_2026	96.7	95
apexAgents	33	47.1
arc_agi_1	93	92.5
ARC-AGI-2	68.8	72.1
arena_elo	1497	1477
arena_text_factuality	1487	1466
Artificial Analysis Coding Index	48.1	70.1
charxiv_reasoning_with_tools	84.7	84.9
charxiv_rq	69.1	84.2
coding_arena_elo	1545	1490
critpt	12.6	13.1
finance_agent	76.7	57.9
frontiermath_tier_4	22.9	14.6
gdpval	55.9	57.8
GDPval-AA (Elo)	1619	1656
GPQA Diamond	91.3	92.2
HiL-Bench	38.3	27.7
HLE	53.1	42.7
hmmt_feb_2026	96.2	95.5
IFBench	62.5	76.3
Legal Agent Benchmark	4.2	0.8
livebench	76.3	75
livebench_agentic_coding	49	49
livebench_coding	78.2	78.2
livebench_data_analysis	69.9	64.9
livebench_instruction_following	63.3	75.6
livebench_language	83.3	84.6
livebench_math	89.3	88.2
livebench_reasoning	88.7	82
matharena_visual_math_overall	72.3	89.9
MCP Atlas	76.8	83.6
MMMU-Pro	77.3	84.3
mrcr_v2_8needle_128k_average	84	77.3
OmniScience Accuracy	47	51.4
OmniScience Non-Hallucination	37.2	38.2
OSWorld-Verified	72.7	78.4
scicode	52	53.1
simplebench	67.6	76.7
simpleqa_verified	46.5	68.4
SWE-bench Pro	57.3	55.1
Terminal-Bench 2.0	65.4	76.2
Terminal-Bench Hard	65.4	46.2
toolathlon	56.8	56.5
τ²-Bench Telecom (AA run)	99.3	95.6
τ³-Bench	72.4	25.4

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.