Qwen3.6-35B-A3B vs Qwen3.8-27B

New Comparison

Scores are fetched live from the current database on every visit — nothing is cached.

Winner

Qwen3.8-27B

Qwen3.8-27B wins this comparison with an overall score of 87 vs 56 for Qwen3.6-35B-A3B, leading 30 of 32 shared benchmarks.

Qwen3.8-27B 87.4 Qwen3.6-35B-A3B 56

Based on 32 of 103 benchmarks scored by all 2 selected models.

Benchmark Scores Comparison

Qwen3.6-35B-A3B Qwen3.8-27B

Qwen3.6-35B-A3B

35.0B params

Score: 100.0

Qwen3.8-27B

27.0B params

Score: 100.0

Coverage matrix

2 models × 103 benchmarks — hover a benchmark name for its category

Full Coverage Matrix
Hides benchmarks where all selected models score identically.
Shading = score relative to the best score among the selected models: Low Mid High Missing (—) The totals row shows how many of the 103 benchmarks each model has scores for.
Benchmark Qwen3.6-35B-A3B Qwen Team (Alibaba Cloud) Qwen3.8-27B Qwen Team (Alibaba Cloud)
HMMT Feb 25 90.7 —
DynaMath 82.8 —
GPQA Diamond 86 89.2
IFBench (prompt loose) — 79.5
MathVista (mini) 86.4 —
MMBench EN-DEV v1.1 92.8 —
LiveCodeBench v6 80.4 90.3
SWE-bench Pro 49.5 61.7
BFCL-V4 64.7 72.4
RealWorldQA 85.3 85.9
Agents' Last Exam (Score) — 42.9
EmbSpatialBench 84.3 —
AA-LCR 69.3 80.3
MathVision — 90
AIME 26 92.7 98.8
AndroidWorld — 81.9
Agents' Last Exam — 20.4
Toolathlon Verified 41.7 67.1
RecreationBench — 47.1
VITA-Bench 35.6 —
TAU3-Bench 67.2 —
AutomationBench (Public) 24.9 58.7
MMStar 80.7 —
GDPVal 67.4 63.5
SWE Atlas - TW 13.3 —
WildClawBench 68.7 —
SWE Atlas - QnA 15.5 —
AI2D_TEST 92.7 —
ParseBench Text Formatting 58.3 —
NL2Repo 29.4 42.3
MMMU-Pro 75.3 —
QwenClawBench 52.6 —
BabyVision — 65.7
Claw-Eval Avg 68.7 56.9
VideoMMMU 83.7 —
HMMT Nov 25 89.1 —
Multi-Challenge 58.2 63.6
SWE-MM — 38.6
MMLU-Pro 85.2 85.7
SuperGPQA 64.7 —
OSWorld-Verified — 84.3
HLE (with tools) 28.9 30.8
LVBench 71.4 72.4
CharXiv (RQ) 78 83.7
GDP.pdf 6.6 11.2
BrowseComp 62 —
MMLU-Redux 93.3 —
NL2Repo-Bench — 42.3
WebArena-Verified — 64.8
ParseBench Text Content 90.7 —
CoWorkBench — 70.7
MCPMark 37 —
AA-Omniscience Index (public set) -7.2 -0.5
SWE-bench Multilingual 67.2 73.8
MLVU 86.2 —
IFBench 66.6 79.5
SWE-bench Verified 73.4 74.6
IMOAnswerBench 78.9 —
Humanity's Last Exam 21.4 30.8
CountBench 96.1 —
ERQA 61.8 65.5
VideoMME (w/o sub.) 82.5 —
Terminal-Bench 2.0 51.5 —
DeepPlanning 25.9 —
QwenSWEBench — 79
QwenWebBench 1397 —
ODInW13 50.8 —
VlmsAreBlind 96.6 —
Terminal-Bench 2.1 (Terminus-2) 52.5 73
DeepSWE 0 —
OmniDocBench 1.5 89.9 91.1
HallusionBench 69.8 —
OCRBench 90 —
HMMT Feb 26 83.6 95.8
Terminal-Bench Lite 63.3 67
CC-OCR 81.9 —
RefCOCO (avg) 92 —
JobBench — 33.4
Artificial Analysis Intelligence Index 32 —
Tool Decathlon 26.9 —
Frontier-Bench v0.1 1.4 —
OSWorld 2.0 (Partial) — 48
ParseBench Mean 44.1 —
OSWorld 2.0 (Binary) — 19.4
ZEROBench_sub 34.4 —
C-Eval 90 —
WideSearch 60.1 —
SimpleVQA 58.9 —
VideoMME (w sub.) 86.6 —
SkillsBench Avg5 28.7 —
Vision2Web — 62.9
Claw-Eval Pass^3 50 57.4
MCP-Atlas 62.8 —
SWE Atlas - RF 11.4 —
V-Star 90.1 —
DeepSWE 1.1 — 42.2
Terminal-Bench 2.1 (Claude Code) 49.2 —
Chatbot Korean (internal) 74.5 90.1
SciCode 35.6 44.6
RefSpatialBench 64.3 —
BrowseComp Plus 34.8 58
MMMU 81.7 —
MVBench 74.6 —
Covered 85/103 50/103

Scorecard

Overall 0–100 score — category strengths (50%), shared-benchmark wins (30%), coverage (20%)

Model Overall score Shared benchmark wins Coverage Strongest at Weakest at
Qwen3.8-27B Winner 87.4 30 of 32 50 of 103 (48.0%) Vision & Language General Agents
Qwen3.6-35B-A3B 56 2 of 32 85 of 103 (82.0%) Vision & Language Multilingual

Category strengths

Where each model stands, per benchmark category — score relative to the best model, adjusted for benchmark coverage. This also ranks models when they share few benchmarks.

Category Qwen3.8-27B Qwen3.6-35B-A3B
Agentic Coding 23 benchmarks 97.8 81
General Agents 23 benchmarks 97.6 52
Vision & Language 15 benchmarks 100 98.5
STEM & Reasoning 9 benchmarks 100 85.2
Document Understanding 8 benchmarks 100 83.2
Spatial Intelligence 6 benchmarks 100 86.3
Video Understanding 6 benchmarks 100 96.8
Knowledge 5 benchmarks 100 91.8
Instruction Following 3 benchmarks 100 75.8
Long Context 2 benchmarks 100 62.9
Multilingual 1 benchmark 100 0
Other 1 benchmark 100 76.7

Fully comparable benchmarks

Scored by all 2 selected models (32)

Benchmark Category Qwen3.6-35B-A3B Qwen3.8-27B
GPQA Diamond stem_reasoning 86 89.2
LiveCodeBench v6 stem_reasoning 80.4 90.3
SWE-bench Pro coding_agent 49.5 61.7
BFCL-V4 general_agent 64.7 72.4
RealWorldQA vision_language 85.3 85.9
AA-LCR long_context 69.3 80.3
AIME 26 stem_reasoning 92.7 98.8
Toolathlon Verified general_agent 41.7 67.1
AutomationBench (Public) general_agent 24.9 58.7
GDPVal general_agent 67.4 63.5
NL2Repo coding_agent 29.4 42.3
Claw-Eval Avg coding_agent 68.7 56.9
Multi-Challenge instruction_following 58.2 63.6
MMLU-Pro knowledge 85.2 85.7
HLE (with tools) stem_reasoning 28.9 30.8
LVBench video_understanding 71.4 72.4
CharXiv (RQ) document_understanding 78 83.7
GDP.pdf long_context 6.6 11.2
AA-Omniscience Index (public set) knowledge -7.2 -0.5
SWE-bench Multilingual coding_agent 67.2 73.8
IFBench instruction_following 66.6 79.5
SWE-bench Verified coding_agent 73.4 74.6
Humanity's Last Exam stem_reasoning 21.4 30.8
ERQA spatial_intelligence 61.8 65.5
Terminal-Bench 2.1 (Terminus-2) coding_agent 52.5 73
OmniDocBench 1.5 document_understanding 89.9 91.1
HMMT Feb 26 stem_reasoning 83.6 95.8
Terminal-Bench Lite coding_agent 63.3 67
Claw-Eval Pass^3 coding_agent 50 57.4
Chatbot Korean (internal) multilingual 74.5 90.1
SciCode — 35.6 44.6
BrowseComp Plus general_agent 34.8 58

Partial coverage benchmarks

Scored by at least one but not all selected models (71) — missing scores show as —

Benchmark Category Qwen3.6-35B-A3B Qwen3.8-27B
HMMT Feb 25 1/2 models stem_reasoning 90.7 —
DynaMath 1/2 models vision_language 82.8 —
IFBench (prompt loose) 1/2 models instruction_following — 79.5
MathVista (mini) 1/2 models vision_language 86.4 —
MMBench EN-DEV v1.1 1/2 models vision_language 92.8 —
Agents' Last Exam (Score) 1/2 models general_agent — 42.9
EmbSpatialBench 1/2 models spatial_intelligence 84.3 —
MathVision 1/2 models vision_language — 90
AndroidWorld 1/2 models general_agent — 81.9
Agents' Last Exam 1/2 models general_agent — 20.4
RecreationBench 1/2 models general_agent — 47.1
VITA-Bench 1/2 models general_agent 35.6 —
TAU3-Bench 1/2 models general_agent 67.2 —
MMStar 1/2 models vision_language 80.7 —
SWE Atlas - TW 1/2 models coding_agent 13.3 —
WildClawBench 1/2 models coding_agent 68.7 —
SWE Atlas - QnA 1/2 models coding_agent 15.5 —
AI2D_TEST 1/2 models document_understanding 92.7 —
ParseBench Text Formatting 1/2 models document_understanding 58.3 —
MMMU-Pro 1/2 models vision_language 75.3 —
QwenClawBench 1/2 models coding_agent 52.6 —
BabyVision 1/2 models vision_language — 65.7
VideoMMMU 1/2 models video_understanding 83.7 —
HMMT Nov 25 1/2 models stem_reasoning 89.1 —
SWE-MM 1/2 models coding_agent — 38.6
SuperGPQA 1/2 models knowledge 64.7 —
OSWorld-Verified 1/2 models general_agent — 84.3
BrowseComp 1/2 models general_agent 62 —
MMLU-Redux 1/2 models knowledge 93.3 —
NL2Repo-Bench 1/2 models coding_agent — 42.3
WebArena-Verified 1/2 models general_agent — 64.8
ParseBench Text Content 1/2 models document_understanding 90.7 —
CoWorkBench 1/2 models general_agent — 70.7
MCPMark 1/2 models general_agent 37 —
MLVU 1/2 models video_understanding 86.2 —
IMOAnswerBench 1/2 models stem_reasoning 78.9 —
CountBench 1/2 models spatial_intelligence 96.1 —
VideoMME (w/o sub.) 1/2 models video_understanding 82.5 —
Terminal-Bench 2.0 1/2 models coding_agent 51.5 —
DeepPlanning 1/2 models general_agent 25.9 —
QwenSWEBench 1/2 models coding_agent — 79
QwenWebBench 1/2 models coding_agent 1397 —
ODInW13 1/2 models spatial_intelligence 50.8 —
VlmsAreBlind 1/2 models vision_language 96.6 —
DeepSWE 1/2 models coding_agent 0 —
HallusionBench 1/2 models vision_language 69.8 —
OCRBench 1/2 models document_understanding 90 —
CC-OCR 1/2 models document_understanding 81.9 —
RefCOCO (avg) 1/2 models spatial_intelligence 92 —
JobBench 1/2 models general_agent — 33.4
Artificial Analysis Intelligence Index 1/2 models composite 32 —
Tool Decathlon 1/2 models general_agent 26.9 —
Frontier-Bench v0.1 1/2 models coding_agent 1.4 —
OSWorld 2.0 (Partial) 1/2 models general_agent — 48
ParseBench Mean 1/2 models document_understanding 44.1 —
OSWorld 2.0 (Binary) 1/2 models general_agent — 19.4
ZEROBench_sub 1/2 models vision_language 34.4 —
C-Eval 1/2 models knowledge 90 —
WideSearch 1/2 models general_agent 60.1 —
SimpleVQA 1/2 models vision_language 58.9 —
VideoMME (w sub.) 1/2 models video_understanding 86.6 —
SkillsBench Avg5 1/2 models coding_agent 28.7 —
Vision2Web 1/2 models vision_language — 62.9
MCP-Atlas 1/2 models general_agent 62.8 —
SWE Atlas - RF 1/2 models coding_agent 11.4 —
V-Star 1/2 models vision_language 90.1 —
DeepSWE 1.1 1/2 models coding_agent — 42.2
Terminal-Bench 2.1 (Claude Code) 1/2 models coding_agent 49.2 —
RefSpatialBench 1/2 models spatial_intelligence 64.3 —
MMMU 1/2 models vision_language 81.7 —
MVBench 1/2 models video_understanding 74.6 —

Comparison FAQ

Which model is better, Qwen3.8-27B or Qwen3.6-35B-A3B?

Qwen3.8-27B wins this comparison with an overall score of 87 vs 56 for Qwen3.6-35B-A3B, leading 30 of 32 shared benchmarks.

Which model is better at Agentic Coding, Qwen3.8-27B or Qwen3.6-35B-A3B?

Qwen3.8-27B leads the Agentic Coding category (strength score 97.8/100 across 8 comparable benchmarks).

How reliable is this comparison verdict?

Based on 32 of 103 benchmarks scored by all 2 selected models.

What now?

You've seen 2 models side by side — keep the loop going.

Share this comparison

The URL is the comparison — copy or share it anywhere. No tracking, no cookies.

Not the right matchup?

Swap one model out — the pre-filled form keeps the rest of the selection.

Swap Qwen3.6-35B-A3B × Swap Qwen3.8-27B ×