← open-weight

Phi-4

Microsoft · USA

MIT-licensed 14B from Microsoft Research. Heavy use of synthetic training data is disclosed; English-primary (thin multilingual coverage). Strongest small-model option for permissive-licence EU deployments.

Licence facts

Licence: MIT
Commercial use: Unrestricted
Derivatives: Allowed
Attribution: Minimal
Parameters: 14B dense
Training data: 9.8T tokens, synthetic + filtered web + books + Q&A (partial)
Context: 16K
Last updated: 2024-Q4

Known risks

English-primary, ~8% multilingual data — weak on EU non-English workloads
Synthetic-data heavy — benchmark performance may overstate real-world generalization

Sources

→ Model card → MIT licence

See also

Azure OpenAI Service· API

Reviewed by Ali Madjaji · Last reviewed 2026-04-15

Home·Methodology·About·/