Which is better, Qwen3.5 122B A10B or Qwen3.5 27B?

Refresh Models —Deals —Regulatory —Sources —

Questions this page answers

Which is better, Qwen3.5 122B A10B or Qwen3.5 27B?: Across 70 shared benchmarks, Qwen3.5 122B A10B scores higher on 40 and Qwen3.5 27B on 28, with 2 level. The widest gap is τ³-Bench, where Qwen3.5 27B scores 68.4 against 13.6. Qwen3.5 27B is the cheaper of the two on tracked API pricing ($0.30 against $0.40 per million input tokens).

Qwen3.5 122B A10B vs Qwen3.5 27B

Across 70 shared benchmarks, Qwen3.5 122B A10B scores higher on 40 and Qwen3.5 27B on 28, with 2 level. The widest gap is τ³-Bench, where Qwen3.5 27B scores 68.4 against 13.6. Qwen3.5 27B is the cheaper of the two on tracked API pricing ($0.30 against $0.40 per million input tokens).

AlibabavsAlibaba70 shared benchmarks40–28 head-to-head

Benchmark	Qwen3.5 122B A10B	Qwen3.5 27B
AA Agentic Index	21.3	54.6
AA Intelligence	32.8	34.6
AA-LCR	70.3	72.3
AA-Omniscience	-41.5	-44
AI2D	93.3	92.9
arena_vision	1246	1241
Artificial Analysis Coding Index	45.7	34.9
baby_vision	40.2	44.6
bfcl_v4	72.2	68.5
browsecomp	63.8	61
browsecomp_zh	69.9	62.1
C-Eval	91.9	90.5
CC-OCR	81.8	81
coding_arena_elo	1358	1357
critpt	0.9	0.9
DeepPlanning	24.1	22.6
DynaMath	85.9	87.7
EmbSpatial-Bench	0.8	84.5
ERQA	62	60.5
gdpval	24.3	33
GPQA Diamond	86.6	85.8
HallusionBench	67.6	70
HLE	47.5	48.5
HMMT 2025	90.3	89.8
IFBench	76.1	76.5
ifeval	93.4	95
include	82.8	81.6
LiveCodeBench v6	78.9	80.7
longbench_v2	60.2	60.6
LVBench	74.4	73.6
mathvision	86.2	86
MathVista	87.4	87.8
MLVU	87.3	85.9
mmlu_prox	82.2	82.2
mmlu_redux	94	93.2
MMLU-Pro	86.7	86.1
mmmlu	86.7	85.9
MMMU	83.9	82.3
MMMU-Pro	76.9	75
MMStar	82.9	81
MMVU	74.7	73.3
multichallenge	61.5	60.8
MV-Bench	76.6	74.6
OCRBench	92.1	89.4
OJBench	39.5	40.1
OmniDocBench 1.5	89.8	88.9
OmniScience Accuracy	24.4	20.7
OmniScience Non-Hallucination	12.9	24.9
OSWorld-Verified	58	56.2
RealWorldQA	85.1	83.7
RefSpatial-Bench	0.7	67.7
scicode	42	39.5
screenspot_pro_no_tools	70.4	70.3
Seal-0	44.1	47.2
SimpleVQA	0.6	56
supergpqa	67.1	65.6
SWE-bench Verified	72	75
t2-bench	79.5	79
Terminal-Bench 2.0	49.4	41.6
Terminal-Bench Hard	31.1	32.6
vectara_answer_rate	99.8	99.8
vectara_avg_summary_length	86.4	94.4
vectara_factual_consistency	88.8	87.9
vectara_hallucination_rate ↓	11.2	12.1
VideoMMMU	82	82.3
VITA-Bench	33.6	41.9
WideSearch	60.5	66.4
ZeroBench	0.1	0.1
τ²-Bench Telecom (AA run)	93.6	93.9
τ³-Bench	13.6	68.4

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