Which is better, Gemma 4 31B or Qwen3.5 122B A10B?

Refresh Models —Deals —Regulatory —Sources —

Questions this page answers

Which is better, Gemma 4 31B or Qwen3.5 122B A10B?: Across 51 shared benchmarks, Gemma 4 31B scores higher on 16 and Qwen3.5 122B A10B on 34, with 1 level. The widest gap is RefSpatial-Bench, where Gemma 4 31B scores 4.7 against 0.7. Gemma 4 31B is the cheaper of the two on tracked API pricing ($0.15 against $0.40 per million input tokens).

Gemma 4 31B vs Qwen3.5 122B A10B

Across 51 shared benchmarks, Gemma 4 31B scores higher on 16 and Qwen3.5 122B A10B on 34, with 1 level. The widest gap is RefSpatial-Bench, where Gemma 4 31B scores 4.7 against 0.7. Gemma 4 31B is the cheaper of the two on tracked API pricing ($0.15 against $0.40 per million input tokens).

GooglevsAlibaba51 shared benchmarks16–34 head-to-head

Benchmark	Gemma 4 31B	Qwen3.5 122B A10B
AA Agentic Index	14.4	21.3
AA Intelligence	29.7	32.8
AA-LCR	68.3	70.3
AA-Omniscience	-47.9	-41.5
AI2D	89	93.3
arena_vision	1275	1246
Artificial Analysis Coding Index	43.4	45.7
C-Eval	82.6	91.9
CC-OCR	75.7	81.8
coding_arena_elo	1364	1358
critpt	1.4	0.9
DeepPlanning	24	24.1
DynaMath	79.5	85.9
ERQA	57.5	62
gdpval	15.5	24.3
GPQA Diamond	85.7	86.6
HallusionBench	67.4	67.6
HLE	26.5	47.5
IFBench	75.6	76.1
LiveCodeBench v6	80	78.9
mathvision	85.6	86.2
MathVista	79.3	87.4
mmlu_redux	93.7	94
MMLU-Pro	85.2	86.7
mmmlu	88.4	86.7
MMMU	80.4	83.9
MMMU-Pro	76.9	76.9
MMStar	77.3	82.9
OCRBench	86.1	92.1
OmniDocBench 1.5	80.1	89.8
OmniScience Accuracy	20	24.4
OmniScience Non-Hallucination	18.1	12.9
RealWorldQA	72.3	85.1
RefSpatial-Bench	4.7	0.7
scicode	43.4	42
SimpleVQA	52.9	0.6
supergpqa	65.7	67.1
SWE-bench Verified	52	72
t2-bench	86.4	79.5
TauBench V3 - Banking	14.8	15.3
Terminal-Bench 2.0	42.9	49.4
Terminal-Bench 2.1	43.4	47.6
Terminal-Bench Hard	36.4	31.1
vectara_answer_rate	100	99.8
vectara_avg_summary_length	75.8	86.4
vectara_factual_consistency	92.6	88.8
vectara_hallucination_rate ↓	7.4	11.2
VITA-Bench	43	33.6
WideSearch	35.2	60.5
τ²-Bench Telecom (AA run)	65.5	93.6
τ³-Bench	67.5	13.6

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. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (direct, direct), otherwise the lowest tracked offer.