Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
o3
65
3/8 categoriesQwen3.5-35B-A3B
67
Winner · 4/8 categorieso3· Qwen3.5-35B-A3B
Pick Qwen3.5-35B-A3B if you want the stronger benchmark profile. o3 only becomes the better choice if agentic is the priority.
Qwen3.5-35B-A3B has the cleaner overall profile here, landing at 67 versus 65. It is a real lead, but still close enough that category-level strengths matter more than the headline number.
Qwen3.5-35B-A3B's sharpest advantage is in coding, where it averages 72.6 against 54.2. The single biggest benchmark swing on the page is LiveCodeBench, 40% to 74.6%. o3 does hit back in agentic, so the answer changes if that is the part of the workload you care about most.
o3 is also the more expensive model on tokens at $10.00 input / $40.00 output per 1M tokens, versus $0.00 input / $0.00 output per 1M tokens for Qwen3.5-35B-A3B. That is roughly Infinityx on output cost alone. Qwen3.5-35B-A3B gives you the larger context window at 262K, compared with 200K for o3.
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 | o3 | Qwen3.5-35B-A3B |
|---|---|---|
| Agentico3 wins | ||
| Terminal-Bench 2.0 | 71% | 40.5% |
| BrowseComp | 75% | 61% |
| OSWorld-Verified | 65% | 54.5% |
| tau2-bench | — | 81.2% |
| CodingQwen3.5-35B-A3B wins | ||
| HumanEval | 78% | — |
| SWE-bench Verified | 71.7% | 69.2% |
| LiveCodeBench | 40% | 74.6% |
| SWE-bench Pro | 58% | — |
| Multimodal & GroundedQwen3.5-35B-A3B wins | ||
| MMMU-Pro | 70% | 75.1% |
| OfficeQA Pro | 75% | — |
| Reasoningo3 wins | ||
| MuSR | 82% | — |
| BBH | 86% | — |
| LongBench v2 | 82% | 59% |
| MRCRv2 | 81% | — |
| ARC-AGI-2 | 3% | — |
| KnowledgeQwen3.5-35B-A3B wins | ||
| MMLU | 86% | — |
| GPQA | 87% | 84.2% |
| SuperGPQA | 85% | 63.4% |
| MMLU-Pro | 75% | 85.3% |
| HLE | 24% | — |
| FrontierScience | 77% | — |
| SimpleQA | 84% | — |
| Instruction FollowingQwen3.5-35B-A3B wins | ||
| IFEval | 85% | 91.9% |
| Multilingualo3 wins | ||
| MGSM | 83% | — |
| MMLU-ProX | 80% | 81% |
| Mathematics | ||
| AIME 2023 | 88% | — |
| AIME 2024 | 90% | — |
| AIME 2025 | 89% | — |
| HMMT Feb 2023 | 84% | — |
| HMMT Feb 2024 | 86% | — |
| HMMT Feb 2025 | 85% | — |
| BRUMO 2025 | 87% | — |
| MATH-500 | 88% | — |
Qwen3.5-35B-A3B is ahead overall, 67 to 65. The biggest single separator in this matchup is LiveCodeBench, where the scores are 40% and 74.6%.
Qwen3.5-35B-A3B has the edge for knowledge tasks in this comparison, averaging 79.3 versus 67.4. 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 54.2. Inside this category, LiveCodeBench is the benchmark that creates the most daylight between them.
o3 has the edge for reasoning in this comparison, averaging 62 versus 59. Inside this category, LongBench v2 is the benchmark that creates the most daylight between them.
o3 has the edge for agentic tasks in this comparison, averaging 69.9 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 72.3. 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 85. Inside this category, IFEval is the benchmark that creates the most daylight between them.
o3 has the edge for multilingual tasks in this comparison, averaging 81.1 versus 81. Inside this category, MMLU-ProX is the benchmark that creates the most daylight between them.
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.