11 models tracked · known for Phi-2, Phi-2, Phi-4
Claims drawn from cited facts, not live model generation.
This lab is Microsoft.
1 cited fact
Microsoft has shipped 11 models tracked in this graph, giving it a substantial portfolio footprint built on repeated model development. No recent blog announcements or papers are tracked for Microsoft alongside those models, so the record here reads as a product-focused shipper rather than a lab whose output is paired with a visible stream of research publication or public commentary.
3 cited facts
47 benchmark scores across this lab’s models, tagged by evidence status.
| Model | Best score | Claim status |
|---|---|---|
| phi-3-medium 14B | 88.8 | independently reproduced |
| phi-3-small 7.4B | 87.6 | self-reported |
| phi-3-mini 3.8B | 84 | self-reported |
| Phi-4 | 79.73 | independently reproduced |
| Phi-2 | 67.87 | self-reported |
| Phi-1.5 | 46.8 | self-reported |
| Phi-3-mini-4k-instruct | — | unverified |
| Phi-3.5-mini-instruct | — | unverified |
| Phi-tiny-MoE-instruct | — | unverified |
Where a model from Microsoft holds the top score, with the benchmark’s integrity grade.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks, huggingface_models