Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
1-bit Bonsai 1.7B
~39
0/8 categoriesDeepSeek V3.2 (Thinking)
68
Winner · 4/8 categories1-bit Bonsai 1.7B· DeepSeek V3.2 (Thinking)
Pick DeepSeek V3.2 (Thinking) if you want the stronger benchmark profile. 1-bit Bonsai 1.7B only becomes the better choice if you would rather avoid the extra latency and token burn of a reasoning model.
DeepSeek V3.2 (Thinking) is clearly ahead on the aggregate, 68 to 39. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
DeepSeek V3.2 (Thinking)'s sharpest advantage is in mathematics, where it averages 86.3 against 34.4. The single biggest benchmark swing on the page is GPQA, 20.7% to 85%.
DeepSeek V3.2 (Thinking) is the reasoning model in the pair, while 1-bit Bonsai 1.7B is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. DeepSeek V3.2 (Thinking) gives you the larger context window at 128K, compared with 32K for 1-bit Bonsai 1.7B.
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 | 1-bit Bonsai 1.7B | DeepSeek V3.2 (Thinking) |
|---|---|---|
| Agentic | ||
| Terminal-Bench 2.0 | — | 71% |
| BrowseComp | — | 70% |
| OSWorld-Verified | — | 67% |
| Coding | ||
| HumanEval | — | 79% |
| SWE-bench Verified | — | 48% |
| LiveCodeBench | — | 45% |
| SWE-bench Pro | — | 58% |
| Multimodal & Grounded | ||
| MMMU-Pro | — | 66% |
| OfficeQA Pro | — | 77% |
| ReasoningDeepSeek V3.2 (Thinking) wins | ||
| MuSR | 45.1% | 81% |
| BBH | — | 86% |
| LongBench v2 | — | 78% |
| MRCRv2 | — | 78% |
| ARC-AGI-2 | — | 4% |
| KnowledgeDeepSeek V3.2 (Thinking) wins | ||
| GPQA | 20.7% | 85% |
| MMLU | — | 87% |
| SuperGPQA | — | 83% |
| MMLU-Pro | — | 73% |
| HLE | — | 22% |
| FrontierScience | — | 77% |
| SimpleQA | — | 83% |
| Instruction FollowingDeepSeek V3.2 (Thinking) wins | ||
| IFEval | 63% | 85% |
| Multilingual | ||
| MGSM | — | 84% |
| MMLU-ProX | — | 79% |
| MathematicsDeepSeek V3.2 (Thinking) wins | ||
| MATH-500 | 34.4% | 84% |
| AIME 2023 | — | 87% |
| AIME 2024 | — | 89% |
| AIME 2025 | — | 88% |
| HMMT Feb 2023 | — | 83% |
| HMMT Feb 2024 | — | 85% |
| HMMT Feb 2025 | — | 84% |
| BRUMO 2025 | — | 86% |
DeepSeek V3.2 (Thinking) is ahead overall, 68 to 39. The biggest single separator in this matchup is GPQA, where the scores are 20.7% and 85%.
DeepSeek V3.2 (Thinking) has the edge for knowledge tasks in this comparison, averaging 65.9 versus 20.7. Inside this category, GPQA is the benchmark that creates the most daylight between them.
DeepSeek V3.2 (Thinking) has the edge for math in this comparison, averaging 86.3 versus 34.4. Inside this category, MATH-500 is the benchmark that creates the most daylight between them.
DeepSeek V3.2 (Thinking) has the edge for reasoning in this comparison, averaging 60.1 versus 45.1. Inside this category, MuSR is the benchmark that creates the most daylight between them.
DeepSeek V3.2 (Thinking) has the edge for instruction following in this comparison, averaging 85 versus 63. Inside this category, IFEval 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.