Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
Holo3-35B-A3B
~78
Winner · 1/8 categoriesMiniMax M2.7
~66
0/8 categoriesHolo3-35B-A3B· MiniMax M2.7
Pick Holo3-35B-A3B if you want the stronger benchmark profile. MiniMax M2.7 only becomes the better choice if you want the cheaper token bill or you need the larger 200K context window.
Holo3-35B-A3B is clearly ahead on the aggregate, 78 to 66. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Holo3-35B-A3B's sharpest advantage is in agentic, where it averages 77.8 against 57.
Holo3-35B-A3B is also the more expensive model on tokens at $0.25 input / $1.80 output per 1M tokens, versus $0.30 input / $1.20 output per 1M tokens for MiniMax M2.7. MiniMax M2.7 gives you the larger context window at 200K, compared with 64K for Holo3-35B-A3B.
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 | Holo3-35B-A3B | MiniMax M2.7 |
|---|---|---|
| AgenticHolo3-35B-A3B wins | ||
| OSWorld-Verified | 77.8% | — |
| Terminal-Bench 2.0 | — | 57% |
| Toolathlon | — | 46.3% |
| MLE-Bench Lite | — | 66.6% |
| MM-ClawBench | — | 62.7% |
| Coding | ||
| SWE-bench Pro | — | 56.2% |
| SWE Multilingual | — | 76.5% |
| Multi-SWE Bench | — | 52.7% |
| VIBE-Pro | — | 55.6% |
| NL2Repo | — | 39.8% |
| Multimodal & Grounded | ||
| GDPval-AA | — | 1495 |
| Reasoning | ||
| Coming soon | ||
| Knowledge | ||
| Coming soon | ||
| Instruction Following | ||
| Coming soon | ||
| Multilingual | ||
| Coming soon | ||
| Mathematics | ||
| Coming soon | ||
Holo3-35B-A3B is ahead overall, 78 to 66.
Holo3-35B-A3B has the edge for agentic tasks in this comparison, averaging 77.8 versus 57. MiniMax M2.7 stays close enough that the answer can still flip depending on your workload.
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.