Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
1-bit Bonsai 4B
~44
Winner · 1/8 categoriesNemotron Ultra 253B
41
3/8 categories1-bit Bonsai 4B· Nemotron Ultra 253B
Pick 1-bit Bonsai 4B if you want the stronger benchmark profile. Nemotron Ultra 253B only becomes the better choice if knowledge is the priority or you want the stronger reasoning-first profile.
1-bit Bonsai 4B has the cleaner overall profile here, landing at 44 versus 41. It is a real lead, but still close enough that category-level strengths matter more than the headline number.
1-bit Bonsai 4B's sharpest advantage is in mathematics, where it averages 65.8 against 55.3. The single biggest benchmark swing on the page is GPQA, 28.7% to 48%. Nemotron Ultra 253B does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.
Nemotron Ultra 253B 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.
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 | Nemotron Ultra 253B |
|---|---|---|
| Agentic | ||
| Terminal-Bench 2.0 | — | 46% |
| BrowseComp | — | 50% |
| OSWorld-Verified | — | 45% |
| Coding | ||
| HumanEval | — | 41% |
| SWE-bench Verified | — | 31% |
| LiveCodeBench | — | 30% |
| SWE-bench Pro | — | 38% |
| Multimodal & Grounded | ||
| MMMU-Pro | — | 37% |
| OfficeQA Pro | — | 54% |
| ReasoningNemotron Ultra 253B wins | ||
| MuSR | 41.4% | 45% |
| BBH | — | 77% |
| LongBench v2 | — | 55% |
| MRCRv2 | — | 56% |
| KnowledgeNemotron Ultra 253B wins | ||
| GPQA | 28.7% | 48% |
| MMLU | — | 49% |
| SuperGPQA | — | 46% |
| MMLU-Pro | — | 63% |
| HLE | — | 11% |
| FrontierScience | — | 49% |
| SimpleQA | — | 47% |
| Instruction FollowingNemotron Ultra 253B wins | ||
| IFEval | 69.6% | 78% |
| Multilingual | ||
| MGSM | — | 74% |
| MMLU-ProX | — | 68% |
| Mathematics1-bit Bonsai 4B wins | ||
| MATH-500 | 65.8% | 74% |
| AIME 2023 | — | 49% |
| AIME 2024 | — | 51% |
| AIME 2025 | — | 50% |
| HMMT Feb 2023 | — | 45% |
| HMMT Feb 2024 | — | 47% |
| HMMT Feb 2025 | — | 46% |
| BRUMO 2025 | — | 48% |
1-bit Bonsai 4B is ahead overall, 44 to 41. The biggest single separator in this matchup is GPQA, where the scores are 28.7% and 48%.
Nemotron Ultra 253B has the edge for knowledge tasks in this comparison, averaging 42.6 versus 28.7. Inside this category, GPQA is the benchmark that creates the most daylight between them.
1-bit Bonsai 4B has the edge for math in this comparison, averaging 65.8 versus 55.3. Inside this category, MATH-500 is the benchmark that creates the most daylight between them.
Nemotron Ultra 253B has the edge for reasoning in this comparison, averaging 52.7 versus 41.4. Inside this category, MuSR is the benchmark that creates the most daylight between them.
Nemotron Ultra 253B has the edge for instruction following in this comparison, averaging 78 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.