Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
1-bit Bonsai 4B
~44
0/8 categoriesGemini 3 Pro Deep Think
80
Winner · 4/8 categories1-bit Bonsai 4B· Gemini 3 Pro Deep Think
Pick Gemini 3 Pro Deep Think if you want the stronger benchmark profile. 1-bit Bonsai 4B only becomes the better choice if you would rather avoid the extra latency and token burn of a reasoning model.
Gemini 3 Pro Deep Think is clearly ahead on the aggregate, 80 to 44. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Gemini 3 Pro Deep Think's sharpest advantage is in knowledge, where it averages 76.4 against 28.7. The single biggest benchmark swing on the page is GPQA, 28.7% to 97%.
Gemini 3 Pro Deep Think is the reasoning model in the pair, while 1-bit Bonsai 4B 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. Gemini 3 Pro Deep Think gives you the larger context window at 2M, compared with 32K for 1-bit Bonsai 4B.
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 4B | Gemini 3 Pro Deep Think |
|---|---|---|
| Agentic | ||
| Terminal-Bench 2.0 | — | 77% |
| BrowseComp | — | 87% |
| OSWorld-Verified | — | 73% |
| Coding | ||
| HumanEval | — | 91% |
| SWE-bench Verified | — | 58% |
| LiveCodeBench | — | 58% |
| SWE-bench Pro | — | 63% |
| Multimodal & Grounded | ||
| MMMU-Pro | — | 95% |
| OfficeQA Pro | — | 95% |
| ReasoningGemini 3 Pro Deep Think wins | ||
| MuSR | 41.4% | 93% |
| BBH | — | 95% |
| LongBench v2 | — | 94% |
| MRCRv2 | — | 96% |
| ARC-AGI-2 | — | 45.1% |
| KnowledgeGemini 3 Pro Deep Think wins | ||
| GPQA | 28.7% | 97% |
| MMLU | — | 99% |
| SuperGPQA | — | 95% |
| MMLU-Pro | — | 81% |
| HLE | — | 32% |
| FrontierScience | — | 88% |
| SimpleQA | — | 95% |
| Instruction FollowingGemini 3 Pro Deep Think wins | ||
| IFEval | 69.6% | 89% |
| Multilingual | ||
| MGSM | — | 92% |
| MMLU-ProX | — | 85% |
| MathematicsGemini 3 Pro Deep Think wins | ||
| MATH-500 | 65.8% | 92% |
| AIME 2023 | — | 99% |
| AIME 2024 | — | 99% |
| AIME 2025 | — | 98% |
| HMMT Feb 2023 | — | 95% |
| HMMT Feb 2024 | — | 97% |
| HMMT Feb 2025 | — | 96% |
| BRUMO 2025 | — | 96% |
Gemini 3 Pro Deep Think is ahead overall, 80 to 44. The biggest single separator in this matchup is GPQA, where the scores are 28.7% and 97%.
Gemini 3 Pro Deep Think has the edge for knowledge tasks in this comparison, averaging 76.4 versus 28.7. Inside this category, GPQA is the benchmark that creates the most daylight between them.
Gemini 3 Pro Deep Think has the edge for math in this comparison, averaging 95.8 versus 65.8. Inside this category, MATH-500 is the benchmark that creates the most daylight between them.
Gemini 3 Pro Deep Think has the edge for reasoning in this comparison, averaging 82.1 versus 41.4. Inside this category, MuSR is the benchmark that creates the most daylight between them.
Gemini 3 Pro Deep Think has the edge for instruction following in this comparison, averaging 89 versus 69.6. 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.