Which is better, Claude Opus 4.6 or Gemma 4 31B?

Across 48 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 42 and Gemma 4 31B leads on 6 — Claude Opus 4.6 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 Gemma 4 31B

AnthropicvsGoogle48 shared benchmarks42–6 head-to-head

Benchmark	Claude Opus 4.6	Gemma 4 31B
AA Agentic Index	67.6	14.4
AA Intelligence	44.9	29.7
AA-LCR	74.3	68.3
AA-Omniscience	13.7	-47.9
aime_2026	96.7	89.2
arena_elo	1497	1451
arena_text_factuality	1487	1449
arena_vision	1300	1252
Artificial Analysis Coding Index	48.1	43.4
charxiv_rq	69.1	67.9
coding_arena_elo	1545	1364
critpt	12.6	1.4
ERQA	40.8	57.5
gdpval	55.9	15.5
GPQA Diamond	91.3	85.7
HLE	53.1	26.5
HLE (with tools)	62.7	26.5
hmmt_feb_2026	96.2	77.2
hmmt_nov_2025	96.3	87.5
IFBench	62.5	75.6
imo_answer_bench	75.3	74.5
LiveCodeBench v6	88.8	80
mathvision	71.2	85.6
MCP Atlas	76.8	57.2
mcpmark	56.7	18.1
MMLU-Pro	89.1	85.2
mmmlu	91.1	88.4
MMMU-Pro	77.3	76.9
mrcr_v2_8needle_128k_average	84	66.4
nl2repo	49.8	15.5
OmniDocBench 1.5	86.6	80.1
OmniScience Accuracy	47	20
OmniScience Non-Hallucination	37.2	18.1
RealWorldQA	73.9	72.3
scicode	52	43.4
simpleqa_verified	46.5	9.6
SWE-bench Multilingual	77.8	51.7
SWE-bench Pro	57.3	35.7
SWE-bench Verified	80.8	52
Terminal-Bench 2.0	65.4	42.9
Terminal-Bench Hard	65.4	36.4
tool_decathlon	47.2	21.2
vectara_answer_rate	99.8	100
vectara_avg_summary_length	137.6	75.8
vectara_factual_consistency	87.8	92.6
vectara_hallucination_rate ↓	12.2	7.4
τ²-Bench Telecom (AA run)	99.3	65.5
τ³-Bench	72.4	67.5

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.