Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
1-bit Bonsai 8B
~50
0/8 categoriesDeepSeekMath V2
61
Winner · 4/8 categories1-bit Bonsai 8B· DeepSeekMath V2
Pick DeepSeekMath V2 if you want the stronger benchmark profile. 1-bit Bonsai 8B only becomes the better choice if you would rather avoid the extra latency and token burn of a reasoning model.
DeepSeekMath V2 is clearly ahead on the aggregate, 61 to 50. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
DeepSeekMath V2's sharpest advantage is in knowledge, where it averages 62.3 against 30. The single biggest benchmark swing on the page is GPQA, 30% to 79%.
DeepSeekMath V2 is the reasoning model in the pair, while 1-bit Bonsai 8B 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. DeepSeekMath V2 gives you the larger context window at 128K, compared with 64K for 1-bit Bonsai 8B.
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 8B | DeepSeekMath V2 |
|---|---|---|
| Agentic | ||
| Terminal-Bench 2.0 | — | 65% |
| BrowseComp | — | 66% |
| OSWorld-Verified | — | 61% |
| Coding | ||
| HumanEval | — | 72% |
| SWE-bench Verified | — | 45% |
| LiveCodeBench | — | 44% |
| SWE-bench Pro | — | 51% |
| Multimodal & Grounded | ||
| MMMU-Pro | — | 64% |
| OfficeQA Pro | — | 73% |
| ReasoningDeepSeekMath V2 wins | ||
| MuSR | 50% | 75% |
| BBH | — | 86% |
| LongBench v2 | — | 75% |
| MRCRv2 | — | 72% |
| KnowledgeDeepSeekMath V2 wins | ||
| GPQA | 30% | 79% |
| MMLU | — | 80% |
| SuperGPQA | — | 77% |
| MMLU-Pro | — | 74% |
| HLE | — | 18% |
| FrontierScience | — | 73% |
| SimpleQA | — | 77% |
| Instruction FollowingDeepSeekMath V2 wins | ||
| IFEval | 79.8% | 83% |
| Multilingual | ||
| MGSM | — | 87% |
| MMLU-ProX | — | 80% |
| MathematicsDeepSeekMath V2 wins | ||
| MATH-500 | 66% | 90% |
| AIME 2023 | — | 80% |
| AIME 2024 | — | 82% |
| AIME 2025 | — | 81% |
| HMMT Feb 2023 | — | 76% |
| HMMT Feb 2024 | — | 78% |
| HMMT Feb 2025 | — | 77% |
| BRUMO 2025 | — | 79% |
DeepSeekMath V2 is ahead overall, 61 to 50. The biggest single separator in this matchup is GPQA, where the scores are 30% and 79%.
DeepSeekMath V2 has the edge for knowledge tasks in this comparison, averaging 62.3 versus 30. Inside this category, GPQA is the benchmark that creates the most daylight between them.
DeepSeekMath V2 has the edge for math in this comparison, averaging 82.6 versus 66. Inside this category, MATH-500 is the benchmark that creates the most daylight between them.
DeepSeekMath V2 has the edge for reasoning in this comparison, averaging 74 versus 50. Inside this category, MuSR is the benchmark that creates the most daylight between them.
DeepSeekMath V2 has the edge for instruction following in this comparison, averaging 83 versus 79.8. 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.