Qwen3.6-35B-A3B vs Qwen3.5-397B-A17B vs Qwen3.5-9B
New ComparisonScores are fetched live from the current database on every visit — nothing is cached.
Winner
Qwen3.5-397B-A17B
Qwen3.5-397B-A17B wins this comparison with an overall score of 84 vs 63 for Qwen3.6-35B-A3B, leading 14 of 17 shared benchmarks.
Qwen3.5-397B-A17B
84.2
Qwen3.6-35B-A3B
63.2
Qwen3.5-9B
24.6
Based on 17 of 120 benchmarks scored by all 3 selected models.
Benchmark Scores Comparison
Qwen3.6-35B-A3B
Qwen3.5-397B-A17B
Qwen3.5-9B
Qwen3.6-35B-A3B
35.0B params
Score: 100.0
Qwen3.5-397B-A17B
Score: 15.0
Qwen3.5-9B
Score: 15.0
Coverage matrix
3 models × 120 benchmarks — hover a benchmark name for its category
Shading = score relative to the best score among the selected models:
Low
Mid
High
Missing (—)
The totals row shows how many of the 120 benchmarks each model has scores for.
| Benchmark | Qwen3.6-35B-A3B Qwen Team (Alibaba Cloud) | Qwen3.5-397B-A17B Qwen Team (Alibaba Cloud) | Qwen3.5-9B Qwen Team (Alibaba Cloud) |
|---|---|---|---|
| OmniScience Accuracy | — | 35.9 | — |
| Apex-Shortlist (with tools) | — | 60.4 | — |
| HMMT Feb 25 | 90.7 | 94.8 | — |
| DynaMath | 82.8 | 86.3 | — |
| GPQA Diamond | 86 | 88.4 | — |
| IFBench (prompt loose) | — | 78.2 | — |
| MathVista (mini) | 86.4 | — | — |
| MMBench EN-DEV v1.1 | 92.8 | — | — |
| ProfBench (Search) | — | 53 | — |
| LiveCodeBench v6 | 80.4 | 83.6 | — |
| TauBench V3 Retail | — | 88.5 | — |
| SWE-bench Pro | 49.5 | 50.9 | 33.8 |
| BFCL-V4 | 64.7 | — | 66.1 |
| RealWorldQA | 85.3 | 83.9 | — |
| EmbSpatialBench | 84.3 | — | — |
| AA-LCR | 69.3 | 68.3 | 63 |
| AIME 26 | 92.7 | 93.3 | 88.2 |
| AgentWorldBench Terminal | — | 55.3 | — |
| Toolathlon Verified | 41.7 | 38.3 | — |
| VITA-Bench | 35.6 | — | 15.6 |
| TAU3-Bench | 67.2 | — | 9.3 |
| AgentWorldBench Overall | — | 54.74 | — |
| OmniScience Non-Hallucination | — | 7.4 | — |
| AutomationBench (Public) | 24.9 | — | — |
| MMStar | 80.7 | 83.8 | — |
| GDPVal | 67.4 | 34.6 | — |
| Apex-Shortlist (no tools) | — | 61.4 | — |
| TauBench V3 Average | — | 71 | — |
| SWE Atlas - TW | 13.3 | 18.5 | 4.4 |
| IOI 2025 | — | 441.3 | — |
| Longbench v2 (≤ 1M) | — | 68.9 | — |
| WildClawBench | 68.7 | 70.7 | — |
| AgentWorldBench Web | — | 48.55 | — |
| SWE Atlas - QnA | 15.5 | 20.4 | 9.2 |
| AI2D_TEST | 92.7 | — | — |
| ParseBench Text Formatting | 58.3 | — | — |
| NL2Repo | 29.4 | 32.2 | 16.2 |
| MMMU-Pro | 75.3 | 79 | — |
| TauBench V3 Airline | — | 76.5 | — |
| QwenClawBench | 52.6 | 51.8 | — |
| Claw-Eval Avg | 68.7 | 70.7 | 53.2 |
| VideoMMMU | 83.7 | 84.7 | — |
| HMMT Nov 25 | 89.1 | 92.7 | — |
| Multi-Challenge | 58.2 | 63.9 | — |
| PinchBench | — | 86.6 | — |
| MMLU-ProX | — | 86.4 | — |
| MMLU-Pro | 85.2 | 87.8 | — |
| SuperGPQA | 64.7 | 70.4 | — |
| AgentWorldBench Search | — | 30.81 | — |
| IFEval | — | — | 91.5 |
| HLE (with tools) | 28.9 | 48.3 | — |
| LVBench | 71.4 | — | — |
| CharXiv (RQ) | 78 | 80.8 | — |
| RULER (1M) | — | 90.1 | — |
| GDP.pdf | 6.6 | — | — |
| Vals.ai Financial Agent 1.1 (without web search) | — | 61.3 | — |
| CritPt (no tools) | — | 2.4 | — |
| BrowseComp | 62 | 40.5 | 8.3 |
| MMLU-Redux | 93.3 | 94.9 | — |
| ParseBench Text Content | 90.7 | — | — |
| MCPMark | 37 | — | 13.4 |
| AA-Omniscience Index (public set) | -7.2 | — | — |
| SWE-bench Multilingual | 67.2 | 69.3 | 43.3 |
| MLVU | 86.2 | 86.7 | — |
| TauBench V3 Telecom | — | 98 | — |
| IFBench | 66.6 | — | 64.5 |
| SWE-bench Verified | 73.4 | 76.2 | 53.1 |
| IMOAnswerBench | 78.9 | 80.9 | 69.8 |
| Humanity's Last Exam | 21.4 | 28.7 | 14.3 |
| CountBench | 96.1 | 97.2 | — |
| ERQA | 61.8 | 67.5 | — |
| VideoMME (w/o sub.) | 82.5 | — | — |
| Terminal-Bench 2.0 | 51.5 | 52.5 | — |
| IMOAnswerBench (with tools) | — | 84.51 | — |
| DeepPlanning | 25.9 | — | — |
| QwenWebBench | 1397 | 1186 | — |
| TAU2-Bench | — | — | 79.1 |
| ODInW13 | 50.8 | — | — |
| VlmsAreBlind | 96.6 | — | — |
| Terminal-Bench 2.1 (Terminus-2) | 52.5 | 53.5 | 21.3 |
| DeepSWE | 0 | 1 | — |
| OmniDocBench 1.5 | 89.9 | — | — |
| HallusionBench | 69.8 | — | — |
| OCRBench | 90 | — | — |
| HMMT Feb 26 | 83.6 | 87.9 | 70.8 |
| Terminal-Bench Lite | 63.3 | — | — |
| AgentWorldBench OS | — | 60.85 | — |
| AgentWorldBench MCP | — | 68.31 | — |
| Vals.ai Financial Agent 1.1 (with web search) | — | 59 | — |
| GPQA | — | — | 77.2 |
| AgentWorldBench SWE | — | 64.44 | — |
| CC-OCR | 81.9 | 82 | — |
| RefCOCO (avg) | 92 | 92.3 | — |
| Artificial Analysis Intelligence Index | 32 | — | — |
| SciCode (subtask) | — | 48 | — |
| Tool Decathlon | 26.9 | — | — |
| Frontier-Bench v0.1 | 1.4 | 1.4 | — |
| ParseBench Mean | 44.1 | — | — |
| ZEROBench_sub | 34.4 | — | — |
| C-Eval | 90 | 93 | — |
| WideSearch | 60.1 | 74 | — |
| SimpleVQA | 58.9 | 67.1 | — |
| VideoMME (w sub.) | 86.6 | 87.5 | — |
| Workspace Bench | — | — | 25.5 |
| SkillsBench Avg5 | 28.7 | 30 | — |
| TauBench V3 Banking | — | 20.9 | — |
| Claw-Eval Pass^3 | 50 | 48.1 | — |
| WMT24++ (en→xx) | — | 86.8 | — |
| MCP-Atlas | 62.8 | 72.3 | 47.4 |
| SWE Atlas - RF | 11.4 | 18.4 | 4.3 |
| V-Star | 90.1 | 95.8 | — |
| AgentWorldBench Android | — | 54.9 | — |
| Terminal-Bench 2.1 (Claude Code) | 49.2 | 48.6 | 18.9 |
| Chatbot Korean (internal) | 74.5 | — | — |
| Gaokao 2026 | — | — | 135.5 |
| SciCode | 35.6 | — | 32.7 |
| RefSpatialBench | 64.3 | — | — |
| BrowseComp Plus | 34.8 | — | — |
| MMMU | 81.7 | 85 | — |
| MVBench | 74.6 | 77.6 | — |
| Covered | 85/120 | 84/120 | 28/120 |
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 |
|---|---|---|---|---|---|
| Qwen3.5-397B-A17B Winner | 84.2 | 14 of 17 | 84 of 120 (70.0%) | Video Understanding | General Agents |
| Qwen3.6-35B-A3B | 63.2 | 3 of 17 | 85 of 120 (71.0%) | Other | Vision & Language |
| Qwen3.5-9B | 24.6 | 0 of 17 | 28 of 120 (23.0%) | Long Context | Agentic Coding |
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 | Qwen3.5-397B-A17B | Qwen3.6-35B-A3B | Qwen3.5-9B |
|---|---|---|---|
| General Agents 31 benchmarks | 54.2 | 93.8 | 35 |
| Agentic Coding 19 benchmarks | 96.4 | 84 | 24.2 |
| STEM & Reasoning 15 benchmarks | 100 | 84.8 | 55.2 |
| Vision & Language 12 benchmarks | 99.5 | 83.7 | — |
| Document Understanding 8 benchmarks | 100 | 90.2 | — |
| Knowledge 7 benchmarks | 100 | 85.8 | — |
| Spatial Intelligence 6 benchmarks | 100 | 87.8 | — |
| Video Understanding 6 benchmarks | 100 | 93.4 | — |
| Instruction Following 4 benchmarks | 77.5 | 86.7 | 73.8 |
| Long Context 4 benchmarks | 98.6 | 100 | 90.9 |
| Other 3 benchmarks | — | 100 | 90.2 |
Fully comparable benchmarks
Scored by all 3 selected models (17)
| Benchmark | Category | Qwen3.6-35B-A3B | Qwen3.5-397B-A17B | Qwen3.5-9B |
|---|---|---|---|---|
| SWE-bench Pro | coding_agent | 49.5 | 50.9 | 33.8 |
| AA-LCR | long_context | 69.3 | 68.3 | 63 |
| AIME 26 | stem_reasoning | 92.7 | 93.3 | 88.2 |
| SWE Atlas - TW | coding_agent | 13.3 | 18.5 | 4.4 |
| SWE Atlas - QnA | coding_agent | 15.5 | 20.4 | 9.2 |
| NL2Repo | coding_agent | 29.4 | 32.2 | 16.2 |
| Claw-Eval Avg | coding_agent | 68.7 | 70.7 | 53.2 |
| BrowseComp | general_agent | 62 | 40.5 | 8.3 |
| SWE-bench Multilingual | coding_agent | 67.2 | 69.3 | 43.3 |
| SWE-bench Verified | coding_agent | 73.4 | 76.2 | 53.1 |
| IMOAnswerBench | stem_reasoning | 78.9 | 80.9 | 69.8 |
| Humanity's Last Exam | stem_reasoning | 21.4 | 28.7 | 14.3 |
| Terminal-Bench 2.1 (Terminus-2) | coding_agent | 52.5 | 53.5 | 21.3 |
| HMMT Feb 26 | stem_reasoning | 83.6 | 87.9 | 70.8 |
| MCP-Atlas | general_agent | 62.8 | 72.3 | 47.4 |
| SWE Atlas - RF | coding_agent | 11.4 | 18.4 | 4.3 |
| Terminal-Bench 2.1 (Claude Code) | coding_agent | 49.2 | 48.6 | 18.9 |
Partial coverage benchmarks
Scored by at least one but not all selected models (103) — missing scores show as —
| Benchmark | Category | Qwen3.6-35B-A3B | Qwen3.5-397B-A17B | Qwen3.5-9B |
|---|---|---|---|---|
| OmniScience Accuracy 1/3 models | knowledge | — | 35.9 | — |
| Apex-Shortlist (with tools) 1/3 models | stem_reasoning | — | 60.4 | — |
| HMMT Feb 25 2/3 models | stem_reasoning | 90.7 | 94.8 | — |
| DynaMath 2/3 models | vision_language | 82.8 | 86.3 | — |
| GPQA Diamond 2/3 models | stem_reasoning | 86 | 88.4 | — |
| IFBench (prompt loose) 1/3 models | instruction_following | — | 78.2 | — |
| MathVista (mini) 1/3 models | vision_language | 86.4 | — | — |
| MMBench EN-DEV v1.1 1/3 models | vision_language | 92.8 | — | — |
| ProfBench (Search) 1/3 models | general_agent | — | 53 | — |
| LiveCodeBench v6 2/3 models | stem_reasoning | 80.4 | 83.6 | — |
| TauBench V3 Retail 1/3 models | general_agent | — | 88.5 | — |
| BFCL-V4 2/3 models | general_agent | 64.7 | — | 66.1 |
| RealWorldQA 2/3 models | vision_language | 85.3 | 83.9 | — |
| EmbSpatialBench 1/3 models | spatial_intelligence | 84.3 | — | — |
| AgentWorldBench Terminal 1/3 models | general_agent | — | 55.3 | — |
| Toolathlon Verified 2/3 models | general_agent | 41.7 | 38.3 | — |
| VITA-Bench 2/3 models | general_agent | 35.6 | — | 15.6 |
| TAU3-Bench 2/3 models | general_agent | 67.2 | — | 9.3 |
| AgentWorldBench Overall 1/3 models | general_agent | — | 54.74 | — |
| OmniScience Non-Hallucination 1/3 models | knowledge | — | 7.4 | — |
| AutomationBench (Public) 1/3 models | general_agent | 24.9 | — | — |
| MMStar 2/3 models | vision_language | 80.7 | 83.8 | — |
| GDPVal 2/3 models | general_agent | 67.4 | 34.6 | — |
| Apex-Shortlist (no tools) 1/3 models | stem_reasoning | — | 61.4 | — |
| TauBench V3 Average 1/3 models | general_agent | — | 71 | — |
| IOI 2025 1/3 models | stem_reasoning | — | 441.3 | — |
| Longbench v2 (≤ 1M) 1/3 models | long_context | — | 68.9 | — |
| WildClawBench 2/3 models | coding_agent | 68.7 | 70.7 | — |
| AgentWorldBench Web 1/3 models | general_agent | — | 48.55 | — |
| AI2D_TEST 1/3 models | document_understanding | 92.7 | — | — |
| ParseBench Text Formatting 1/3 models | document_understanding | 58.3 | — | — |
| MMMU-Pro 2/3 models | vision_language | 75.3 | 79 | — |
| TauBench V3 Airline 1/3 models | general_agent | — | 76.5 | — |
| QwenClawBench 2/3 models | coding_agent | 52.6 | 51.8 | — |
| VideoMMMU 2/3 models | video_understanding | 83.7 | 84.7 | — |
| HMMT Nov 25 2/3 models | stem_reasoning | 89.1 | 92.7 | — |
| Multi-Challenge 2/3 models | instruction_following | 58.2 | 63.9 | — |
| PinchBench 1/3 models | general_agent | — | 86.6 | — |
| MMLU-ProX 1/3 models | multilingual | — | 86.4 | — |
| MMLU-Pro 2/3 models | knowledge | 85.2 | 87.8 | — |
| SuperGPQA 2/3 models | knowledge | 64.7 | 70.4 | — |
| AgentWorldBench Search 1/3 models | general_agent | — | 30.81 | — |
| IFEval 1/3 models | instruction_following | — | — | 91.5 |
| HLE (with tools) 2/3 models | stem_reasoning | 28.9 | 48.3 | — |
| LVBench 1/3 models | video_understanding | 71.4 | — | — |
| CharXiv (RQ) 2/3 models | document_understanding | 78 | 80.8 | — |
| RULER (1M) 1/3 models | long_context | — | 90.1 | — |
| GDP.pdf 1/3 models | long_context | 6.6 | — | — |
| Vals.ai Financial Agent 1.1 (without web search) 1/3 models | general_agent | — | 61.3 | — |
| CritPt (no tools) 1/3 models | stem_reasoning | — | 2.4 | — |
| MMLU-Redux 2/3 models | knowledge | 93.3 | 94.9 | — |
| ParseBench Text Content 1/3 models | document_understanding | 90.7 | — | — |
| MCPMark 2/3 models | general_agent | 37 | — | 13.4 |
| AA-Omniscience Index (public set) 1/3 models | knowledge | -7.2 | — | — |
| MLVU 2/3 models | video_understanding | 86.2 | 86.7 | — |
| TauBench V3 Telecom 1/3 models | general_agent | — | 98 | — |
| IFBench 2/3 models | instruction_following | 66.6 | — | 64.5 |
| CountBench 2/3 models | spatial_intelligence | 96.1 | 97.2 | — |
| ERQA 2/3 models | spatial_intelligence | 61.8 | 67.5 | — |
| VideoMME (w/o sub.) 1/3 models | video_understanding | 82.5 | — | — |
| Terminal-Bench 2.0 2/3 models | coding_agent | 51.5 | 52.5 | — |
| IMOAnswerBench (with tools) 1/3 models | stem_reasoning | — | 84.51 | — |
| DeepPlanning 1/3 models | general_agent | 25.9 | — | — |
| QwenWebBench 2/3 models | coding_agent | 1397 | 1186 | — |
| TAU2-Bench 1/3 models | general_agent | — | — | 79.1 |
| ODInW13 1/3 models | spatial_intelligence | 50.8 | — | — |
| VlmsAreBlind 1/3 models | vision_language | 96.6 | — | — |
| DeepSWE 2/3 models | coding_agent | 0 | 1 | — |
| OmniDocBench 1.5 1/3 models | document_understanding | 89.9 | — | — |
| HallusionBench 1/3 models | vision_language | 69.8 | — | — |
| OCRBench 1/3 models | document_understanding | 90 | — | — |
| Terminal-Bench Lite 1/3 models | coding_agent | 63.3 | — | — |
| AgentWorldBench OS 1/3 models | general_agent | — | 60.85 | — |
| AgentWorldBench MCP 1/3 models | general_agent | — | 68.31 | — |
| Vals.ai Financial Agent 1.1 (with web search) 1/3 models | general_agent | — | 59 | — |
| GPQA 1/3 models | reasoning | — | — | 77.2 |
| AgentWorldBench SWE 1/3 models | general_agent | — | 64.44 | — |
| CC-OCR 2/3 models | document_understanding | 81.9 | 82 | — |
| RefCOCO (avg) 2/3 models | spatial_intelligence | 92 | 92.3 | — |
| Artificial Analysis Intelligence Index 1/3 models | composite | 32 | — | — |
| SciCode (subtask) 1/3 models | stem_reasoning | — | 48 | — |
| Tool Decathlon 1/3 models | general_agent | 26.9 | — | — |
| Frontier-Bench v0.1 2/3 models | coding_agent | 1.4 | 1.4 | — |
| ParseBench Mean 1/3 models | document_understanding | 44.1 | — | — |
| ZEROBench_sub 1/3 models | vision_language | 34.4 | — | — |
| C-Eval 2/3 models | knowledge | 90 | 93 | — |
| WideSearch 2/3 models | general_agent | 60.1 | 74 | — |
| SimpleVQA 2/3 models | vision_language | 58.9 | 67.1 | — |
| VideoMME (w sub.) 2/3 models | video_understanding | 86.6 | 87.5 | — |
| Workspace Bench 1/3 models | — | — | — | 25.5 |
| SkillsBench Avg5 2/3 models | coding_agent | 28.7 | 30 | — |
| TauBench V3 Banking 1/3 models | general_agent | — | 20.9 | — |
| Claw-Eval Pass^3 2/3 models | coding_agent | 50 | 48.1 | — |
| WMT24++ (en→xx) 1/3 models | multilingual | — | 86.8 | — |
| V-Star 2/3 models | vision_language | 90.1 | 95.8 | — |
| AgentWorldBench Android 1/3 models | general_agent | — | 54.9 | — |
| Chatbot Korean (internal) 1/3 models | multilingual | 74.5 | — | — |
| Gaokao 2026 1/3 models | — | — | — | 135.5 |
| SciCode 2/3 models | — | 35.6 | — | 32.7 |
| RefSpatialBench 1/3 models | spatial_intelligence | 64.3 | — | — |
| BrowseComp Plus 1/3 models | general_agent | 34.8 | — | — |
| MMMU 2/3 models | vision_language | 81.7 | 85 | — |
| MVBench 2/3 models | video_understanding | 74.6 | 77.6 | — |
Comparison FAQ
Which model scores highest in this comparison?
Qwen3.5-397B-A17B wins this comparison with an overall score of 84 vs 63 for Qwen3.6-35B-A3B, leading 14 of 17 shared benchmarks.
Which model is strongest at Agentic Coding?
Qwen3.5-397B-A17B leads the Agentic Coding category (strength score 96.4/100 across 17 comparable benchmarks).
How reliable is this comparison verdict?
Based on 17 of 120 benchmarks scored by all 3 selected models.