Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
Holo3-122B-A10B
~79
Winner · 1/8 categoriesKimi K2.5 (Reasoning)
76
0/8 categoriesHolo3-122B-A10B· Kimi K2.5 (Reasoning)
Pick Holo3-122B-A10B if you want the stronger benchmark profile. Kimi K2.5 (Reasoning) only becomes the better choice if you need the larger 128K context window or you want the stronger reasoning-first profile.
Holo3-122B-A10B has the cleaner overall profile here, landing at 79 versus 76. It is a real lead, but still close enough that category-level strengths matter more than the headline number.
Holo3-122B-A10B's sharpest advantage is in agentic, where it averages 78.9 against 57.6. The single biggest benchmark swing on the page is OSWorld-Verified, 78.8% to 63.3%.
Kimi K2.5 (Reasoning) is the reasoning model in the pair, while Holo3-122B-A10B 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. Kimi K2.5 (Reasoning) gives you the larger context window at 128K, compared with 64K for Holo3-122B-A10B.
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 | Holo3-122B-A10B | Kimi K2.5 (Reasoning) |
|---|---|---|
| AgenticHolo3-122B-A10B wins | ||
| OSWorld-Verified | 78.8% | 63.3% |
| Terminal-Bench 2.0 | — | 50.8% |
| BrowseComp | — | 60.6% |
| Coding | ||
| HumanEval | — | 99% |
| SWE-bench Verified | — | 76.8% |
| LiveCodeBench | — | 85% |
| SWE-bench Pro | — | 70% |
| SWE-Rebench | — | 57.4% |
| Multimodal & Grounded | ||
| MMMU-Pro | — | 78.5% |
| OfficeQA Pro | — | 77% |
| Reasoning | ||
| MuSR | — | 86% |
| BBH | — | 91% |
| LongBench v2 | — | 61% |
| MRCRv2 | — | 81% |
| Knowledge | ||
| MMLU | — | 92% |
| GPQA | — | 87.6% |
| SuperGPQA | — | 88% |
| MMLU-Pro | — | 87.1% |
| HLE | — | 27% |
| FrontierScience | — | 80% |
| SimpleQA | — | 54% |
| Instruction Following | ||
| IFEval | — | 94% |
| Multilingual | ||
| MGSM | — | 96% |
| MMLU-ProX | — | 86% |
| Mathematics | ||
| AIME 2023 | — | 94% |
| AIME 2024 | — | 96% |
| AIME 2025 | — | 96.1% |
| HMMT Feb 2023 | — | 90% |
| HMMT Feb 2024 | — | 92% |
| HMMT Feb 2025 | — | 95.4% |
| BRUMO 2025 | — | 93% |
| MATH-500 | — | 92% |
Holo3-122B-A10B is ahead overall, 79 to 76. The biggest single separator in this matchup is OSWorld-Verified, where the scores are 78.8% and 63.3%.
Holo3-122B-A10B has the edge for agentic tasks in this comparison, averaging 78.9 versus 57.6. Inside this category, OSWorld-Verified 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.