Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
LFM2.5-350M
~39
0/8 categoriesPhi-4
40
Winner · 1/8 categoriesLFM2.5-350M· Phi-4
Pick Phi-4 if you want the stronger benchmark profile. LFM2.5-350M only becomes the better choice if you need the larger 32K context window.
Phi-4 finishes one point ahead overall, 40 to 39. That is enough to call, but not enough to treat as a blowout. This matchup comes down to a few meaningful edges rather than one model dominating the board.
Phi-4's sharpest advantage is in knowledge, where it averages 53.6 against 23.8. The single biggest benchmark swing on the page is GPQA, 30.6% to 56.1%.
LFM2.5-350M gives you the larger context window at 32K, compared with 16K for Phi-4.
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 | LFM2.5-350M | Phi-4 |
|---|---|---|
| Agentic | ||
| Terminal-Bench 2.0 | — | 44% |
| BrowseComp | — | 35% |
| OSWorld-Verified | — | 34% |
| Coding | ||
| HumanEval | — | 82.6% |
| SWE-bench Pro | — | 55% |
| Multimodal & Grounded | ||
| MMMU-Pro | — | 54% |
| OfficeQA Pro | — | 38% |
| Reasoning | ||
| LongBench v2 | — | 30% |
| MRCRv2 | — | 33% |
| KnowledgePhi-4 wins | ||
| GPQA | 30.6% | 56.1% |
| MMLU-Pro | 20.0% | — |
| MMLU | — | 84.8% |
| FrontierScience | — | 52% |
| Instruction Following | ||
| IFEval | 77.0% | — |
| Multilingual | ||
| MGSM | — | 80.6% |
| MMLU-ProX | — | 60% |
| Mathematics | ||
| MATH-500 | — | 94.6% |
Phi-4 is ahead overall, 40 to 39. The biggest single separator in this matchup is GPQA, where the scores are 30.6% and 56.1%.
Phi-4 has the edge for knowledge tasks in this comparison, averaging 53.6 versus 23.8. Inside this category, GPQA 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.