Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
o4-mini (high)
58
2/8 categoriesQwen3.5-35B-A3B
67
Winner · 5/8 categorieso4-mini (high)· Qwen3.5-35B-A3B
Pick Qwen3.5-35B-A3B if you want the stronger benchmark profile. o4-mini (high) only becomes the better choice if agentic is the priority.
Qwen3.5-35B-A3B is clearly ahead on the aggregate, 67 to 58. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Qwen3.5-35B-A3B's sharpest advantage is in coding, where it averages 72.6 against 44.9. The single biggest benchmark swing on the page is LiveCodeBench, 34% to 74.6%. o4-mini (high) does hit back in agentic, so the answer changes if that is the part of the workload you care about most.
Qwen3.5-35B-A3B gives you the larger context window at 262K, compared with 200K for o4-mini (high).
BenchLM keeps the benchmark table and the operator tradeoffs on the same page so a better score does not hide a materially slower, pricier, or smaller-context model.
Runtime metrics show N/A when BenchLM does not have a sourced snapshot for that exact model. The scoring rules and freshness policy are documented on the methodology page.
| Benchmark | o4-mini (high) | Qwen3.5-35B-A3B |
|---|---|---|
| Agentico4-mini (high) wins | ||
| Terminal-Bench 2.0 | 58% | 40.5% |
| BrowseComp | 64% | 61% |
| OSWorld-Verified | 55% | 54.5% |
| tau2-bench | — | 81.2% |
| CodingQwen3.5-35B-A3B wins | ||
| HumanEval | 74% | — |
| SWE-bench Verified | 68.1% | 69.2% |
| LiveCodeBench | 34% | 74.6% |
| SWE-bench Pro | 42% | — |
| Multimodal & GroundedQwen3.5-35B-A3B wins | ||
| MMMU-Pro | 66% | 75.1% |
| OfficeQA Pro | 71% | — |
| ReasoningQwen3.5-35B-A3B wins | ||
| MuSR | 78% | — |
| BBH | 83% | — |
| LongBench v2 | 75% | 59% |
| MRCRv2 | 74% | — |
| ARC-AGI-2 | 2.4% | — |
| KnowledgeQwen3.5-35B-A3B wins | ||
| MMLU | 82% | — |
| GPQA | 82% | 84.2% |
| SuperGPQA | 80% | 63.4% |
| MMLU-Pro | 76% | 85.3% |
| HLE | 13% | — |
| FrontierScience | 73% | — |
| SimpleQA | 80% | — |
| Instruction FollowingQwen3.5-35B-A3B wins | ||
| IFEval | 83% | 91.9% |
| Multilingualo4-mini (high) wins | ||
| MGSM | 83% | — |
| MMLU-ProX | 81% | 81% |
| Mathematics | ||
| AIME 2023 | 83% | — |
| AIME 2024 | 93.4% | — |
| AIME 2025 | 92.7% | — |
| HMMT Feb 2023 | 79% | — |
| HMMT Feb 2024 | 81% | — |
| HMMT Feb 2025 | 80% | — |
| BRUMO 2025 | 82% | — |
| MATH-500 | 84% | — |
Qwen3.5-35B-A3B is ahead overall, 67 to 58. The biggest single separator in this matchup is LiveCodeBench, where the scores are 34% and 74.6%.
Qwen3.5-35B-A3B has the edge for knowledge tasks in this comparison, averaging 79.3 versus 62.7. Inside this category, SuperGPQA is the benchmark that creates the most daylight between them.
Qwen3.5-35B-A3B has the edge for coding in this comparison, averaging 72.6 versus 44.9. Inside this category, LiveCodeBench is the benchmark that creates the most daylight between them.
Qwen3.5-35B-A3B has the edge for reasoning in this comparison, averaging 59 versus 57.2. Inside this category, LongBench v2 is the benchmark that creates the most daylight between them.
o4-mini (high) has the edge for agentic tasks in this comparison, averaging 58.5 versus 50.5. Inside this category, Terminal-Bench 2.0 is the benchmark that creates the most daylight between them.
Qwen3.5-35B-A3B has the edge for multimodal and grounded tasks in this comparison, averaging 75.1 versus 68.3. Inside this category, MMMU-Pro is the benchmark that creates the most daylight between them.
Qwen3.5-35B-A3B has the edge for instruction following in this comparison, averaging 91.9 versus 83. Inside this category, IFEval is the benchmark that creates the most daylight between them.
o4-mini (high) has the edge for multilingual tasks in this comparison, averaging 81.7 versus 81. Qwen3.5-35B-A3B stays close enough that the answer can still flip depending on your workload.
Get notified when new models drop, benchmark scores change, or the leaderboard shifts. One email per week.
Free. No spam. Unsubscribe anytime. We only store derived location metadata for consent routing.